Remote sensing image classification labeling method based on color space conversion

Abstract

The present application relates to the field of remote sensing image classification annotation, especially to a remote sensing image classification annotation method based on color space conversion. The content includes: performing color space conversion on the remote sensing image, constructing a ring-shaped color tone enhancement HSV feature vector and a normalized Lab perceptual color feature vector, and calculating a double color space consistency difference index, and obtaining an optimal automatic segmentation threshold value using the maximum inter-class variance method; generating a stable core pixel discrimination map based on the threshold value, constructing a stable candidate ground object region set and a to-be-filled boundary pixel set; constructing a region color dominant feature vector of the stable candidate ground object region, combining with the ground object color prototype to obtain the ground object category of the stable candidate ground object region, and combining with the adjacent stable candidate ground object region to determine the ground object category of the to-be-filled boundary pixel. The problems of misclassification of ground objects, increase of pseudo-boundaries, region fragmentation, and cavities, fractures and unlabeling caused by weak processing of boundary transition zones under complex scenes such as light, shadow and complex scenes are solved.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image classification and annotation, and more particularly to a remote sensing image classification and annotation method based on color space conversion. Background Technology

[0002] With the rapid development of satellite remote sensing, aerial remote sensing, and UAV remote sensing technologies, remote sensing images have been widely used in land use surveys, ecological environment monitoring, disaster identification, and urban planning. However, due to variations in lighting conditions, imaging angles, and sensor responses, different ground features in remote sensing images are prone to color mixing, blurred boundaries, and heterogeneous phenomena. Traditional RGB color space-based classification methods struggle to reliably distinguish complex ground feature regions, resulting in insufficient classification accuracy and labeling consistency. Furthermore, some existing methods rely excessively on manual experience to set thresholds or a large number of manually labeled samples, leading to weak generalization capabilities in various remote sensing environments and failing to meet the demands of high-precision automated remote sensing interpretation.

[0003] Color space conversion methods map remote sensing images from the RGB color space to HSV, Lab, and other color spaces that better align with human visual perception. This enhances the ability to express color differences between different land features, improves the stability of regional boundary identification, and provides a new technical path for automatic classification and refined annotation of remote sensing images. In the future, color space conversion-based remote sensing image classification and annotation methods will be further integrated with multispectral remote sensing, deep learning, and intelligent interpretation technologies, developing towards high-precision, adaptive, low-manual-dependency, and real-time remote sensing applications. Summary of the Invention

[0004] This invention provides a remote sensing image classification and annotation method based on color space conversion to solve the problems of traditional classification methods, which are prone to misclassification of ground features, increased false boundaries, and regional fragmentation under lighting changes, shadow interference, and complex ground feature scenes; weak processing ability for boundary transition areas and mixed pixels, which easily produces boundary holes, regional breaks, and unlabeled areas, affecting the integrity and accuracy of remote sensing image classification and annotation results.

[0005] The remote sensing image classification and annotation method based on color space conversion of the present invention specifically includes the following steps:

[0006] S1. Acquire the remote sensing image to be processed, convert the image from RGB color space to HSV and Lab color spaces, and perform normalization processing to construct a ring-shaped hue-enhanced HSV feature vector and a normalized Lab perceptual color feature vector; based on the ring-shaped hue-enhanced HSV feature vector and the normalized Lab perceptual color feature vector, calculate the dual color space consistency difference index, and use the maximum inter-class variance method to obtain the optimal automatic segmentation threshold; based on the optimal automatic segmentation threshold, generate a stable core pixel discrimination map, and construct a stable candidate land cover region set and a boundary pixel set to be backfilled.

[0007] S2. Construct the regional color dominant feature vector of stable candidate land cover area through regional statistical method, and calculate the color matching difference between stable candidate land cover area and all land cover color prototypes in combination with land cover color prototypes to determine the land cover category of stable candidate land cover area; define neighboring stable candidate land cover areas, calculate the color belonging difference between the boundary pixel to be backfilled and the neighboring stable candidate land cover area, and determine the land cover category of the boundary pixel to be backfilled.

[0008] Preferably, S1 specifically includes:

[0009] We introduce the HSV color space and the Lab color space, and extract the hue component, saturation component and lightness component of the HSV color space, as well as the perceived lightness component, red-green complementary color component and yellow-blue complementary color component of the Lab color space.

[0010] Preferably, S1 specifically includes:

[0011] Linear normalization is performed on the saturation component, brightness component, perceived brightness component, red-green contrasting color component, and yellow-blue contrasting color component to obtain normalized saturation features, normalized brightness features, normalized perceived brightness features, normalized red-green contrasting color features, and normalized yellow-blue contrasting color features. The hue component is converted into a circular vector representation to generate a circular hue enhancement feature. Based on the circular hue enhancement feature, normalized saturation feature, and normalized brightness feature, a circular hue enhancement HSV feature vector is constructed. Based on the normalized perceived brightness feature, normalized red-green contrasting color feature, and normalized yellow-blue contrasting color feature, a normalized Lab perceived color feature vector is constructed.

[0012] Preferably, S1 specifically includes:

[0013] Based on the HSV feature vector enhanced by ring tone and the normalized Lab perceptual color feature vector, using the current pixel and its eight neighboring pixels as the local analysis range, the feature differences between the current pixel and its neighboring pixels in the HSV color space and in the Lab color space are calculated. The feature differences in the HSV color space and the Lab color space are coupled to calculate the dual-color space consistency difference index. The specific calculation formula is as follows:

[0014] in, Represents pixels The consistency difference index of the two color spaces; Represents pixels The set of eight neighboring pixels; Indicates the number of neighboring pixels; Represents pixels The neighboring pixel positions; Indicates Euclidean distance; Represents pixels The HSV feature vector with ring-shaped hue enhancement; Represents pixels The HSV feature vector with ring-shaped hue enhancement; Represents pixels Normalized Lab-sensory color feature vector; Represents pixels Normalized Lab-sensory color feature vector; This indicates the characteristic differences in the HSV color space; This indicates the characteristic differences in the Lab color space.

[0015] Preferably, S1 specifically includes:

[0016] The distribution range of the consistency difference index in the dual color space is statistically analyzed using histograms. A candidate threshold set is constructed by iterating through the minimum and maximum difference indices. Based on the consistency difference index and candidate thresholds, low-difference pixel groups and high-difference pixel groups are constructed. The maximum inter-class variance method is used to select the threshold that maximizes the separation between the low-difference and high-difference pixel groups as the optimal automatic segmentation threshold. Based on the optimal automatic segmentation threshold, a stable core pixel discrimination map is generated. Spatial connectivity search is performed on pixels belonging to the core regions of land features with stable color structures to construct a stable candidate land feature region set. Pixels not belonging to the core regions of land features with stable color structures are designated as boundary pixels to be backfilled, and a boundary pixel set to be backfilled is constructed.

[0017] Preferably, S2 specifically includes:

[0018] Typical land cover category sample areas are extracted from the manually labeled remote sensing image samples. For each land cover category, the sample areas are converted to HSV color space and Lab color space respectively, and the ring hue enhancement feature, normalized saturation feature, normalized brightness feature, normalized perceived brightness feature, normalized red-green contrast color feature and normalized yellow-blue contrast color feature are extracted according to the color feature expression method in step S1, and a sample color feature vector is constructed. The sample color feature vectors of all sample areas of the same land cover category are statistically averaged to obtain the standard color prototype vector corresponding to the land cover category.

[0019] Preferably, S2 specifically includes:

[0020] Based on the regional color dominant feature vector of stable candidate land cover areas and the standard color prototype vectors corresponding to all land cover categories, the color matching difference is calculated, and the land cover category of stable candidate land cover areas is determined by combining the principle of minimum difference.

[0021] Preferably, S2 specifically includes:

[0022] Based on the ring-shaped hue-enhanced HSV feature vector and the normalized Lab-aware color feature vector of the boundary pixels to be backfilled, a color feature vector of the boundary pixels to be backfilled is constructed. Based on the color feature vector of the boundary pixels to be backfilled and the regional color dominant feature vector of the neighboring stable candidate land cover area, the color assignment difference is calculated. The boundary pixels to be backfilled are assigned to the neighboring stable candidate land cover area with the smallest color assignment difference and inherit the land cover category of the neighboring stable candidate land cover area.

[0023] The beneficial effects of the technical solution of the present invention are:

[0024] 1. By simultaneously introducing the HSV and Lab color spaces, a joint expression of color attributes and visual perception differences in remote sensing images is achieved. The HSV color space enhances the ability to distinguish the main color direction and color purity of ground features, while the Lab color space enhances the ability to express color differences perceived by the human eye. This effectively improves the color separability among typical ground features such as vegetation, water bodies, bare land, and buildings, reduces the category confusion problem caused by brightness coupling in the traditional RGB color space, and improves the stability and accuracy of ground feature classification and labeling in complex remote sensing scenarios.

[0025] 2. By constructing a ring-shaped tone enhancement feature and adopting a dual-color space consistency difference index, the distinction between real and false boundaries of ground features is achieved. The ring-shaped tone expression avoids the problem of abrupt distortion at periodic boundaries caused by traditional tone angles. The dual-color space consistency coupling mechanism is used to ensure that the boundary response is enhanced only when there is a significant difference between the HSV and Lab color spaces. This effectively suppresses false detection caused by shadows, local noise, illumination changes, and unstable factors in a single color space, thereby improving the reliability of remote sensing image boundary detection and the stability of region division.

[0026] 3. By employing the maximum inter-class variance method to determine the optimal automatic segmentation threshold and generating a stable core pixel discriminant map, a stable candidate land cover region is constructed. This achieves adaptive threshold separation capability in the remote sensing image region segmentation process. Without the need for manual experience to set a fixed threshold, it can automatically extract regions with stable color structures based on the current remote sensing scene. This enhances the algorithm's adaptability to different remote sensing scenes, imaging conditions, and land cover distributions, reduces human parameter intervention, and improves the engineering deployability and scene generalization capability of the remote sensing image classification and annotation method.

[0027] 4. By constructing a regional color-dominant feature vector and combining it with a land cover color prototype library for regional land cover category matching, and by backfilling the color attribution of boundary pixels under the constraint of nearby stable candidate land cover regions, the integrated processing of regional subject classification and fine boundary completion is realized. This effectively alleviates the problems that are easy to occur in traditional remote sensing image classification methods, such as boundary holes, regional breaks, unlabeled edges, and discontinuous classification in transitional areas, and improves the spatial continuity, boundary integrity, and overall semantic consistency of remote sensing image classification results. Attached Figure Description

[0028] Figure 1 This is a flowchart of the remote sensing image classification and annotation method based on color space conversion described in this invention. Detailed Implementation

[0029] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments; based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0031] The specific scheme of the remote sensing image classification and annotation method based on color space conversion provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0032] See attached document Figure 1 The diagram illustrates a flowchart of a remote sensing image classification and annotation method based on color space conversion according to an embodiment of the present invention. The method includes the following steps:

[0033] S1. Acquire the remote sensing image to be processed, convert the image from RGB color space to HSV and Lab color spaces, and perform normalization processing to construct a ring-shaped hue-enhanced HSV feature vector and a normalized Lab perceptual color feature vector; based on the ring-shaped hue-enhanced HSV feature vector and the normalized Lab perceptual color feature vector, calculate the dual color space consistency difference index, and use the maximum inter-class variance method to obtain the optimal automatic segmentation threshold; based on the optimal automatic segmentation threshold, generate a stable core pixel discrimination map, and construct a stable candidate land cover region set and a boundary pixel set to be backfilled.

[0034] The remote sensing images to be processed are acquired through satellite remote sensing platforms, airborne remote sensing platforms, or UAV remote sensing platforms; each pixel in the remote sensing image contains pixel values ​​for the three visible light channels: red, green, and blue.

[0035] To enhance the color differentiation between different ground features, instead of directly dividing regions within the RGB color space of the remote sensing image, both the HSV and Lab color spaces are introduced. The HSV color space can more directly describe the dominant color direction, color purity, and brightness. The hue component describes the dominant color attribute, the saturation component reflects color intensity, and the brightness component reflects the changes in brightness under visible light conditions. The Lab color space is closer to the human visual perception mechanism. The perceived brightness component reflects perceived brightness, and the complementary color component reflects color changes in the red-green and yellow-blue directions. The HSV and Lab color spaces describe the color characteristics of ground features from two perspectives: color attribute expression and perceived color difference expression, respectively.

[0036] The specific methods for color space conversion are as follows:

[0037] The standard RGB-HSV color space conversion method is used to convert the remote sensing image to be processed from the RGB color space to the HSV color space;

[0038] The CIELAB color space conversion method based on the CIE 1931 XYZ standard color space is used to convert the remote sensing image to be processed from the RGB color space to the Lab color space. Specifically, the standard RGB-XYZ linear color space conversion method is first used to convert the remote sensing image to be processed from the RGB color space to the XYZ color space, and then the CIELAB nonlinear sensing color conversion method is used to further convert the XYZ color space to the Lab color space.

[0039] Because the value ranges of components differ significantly in different color spaces, direct joint calculation can easily lead to one color component dominating the calculation. Therefore, it is necessary to perform linear normalization on color components other than hue using methods such as max-min normalization. The hue component is not processed using ordinary linear normalization because hue is a periodic angular variable, essentially a directional position on the color ring. Therefore, this invention converts the hue component into a circular vector representation, using the directional coordinates of the hue on the unit circle to represent the color direction, thereby maintaining the periodic continuity of the hue and preventing abrupt distortion in the boundary region, which better conforms to the actual color structure change law.

[0040] After color space conversion and normalization, HSV feature vectors with ring-shaped hue enhancement are constructed respectively. and normalized Lab-sensory color feature vector ,in, Represents the pixels in the remote sensing image to be processed Hue components in the HSV color space; Represents pixels Hue components in the HSV color space The projection component along the horizontal axis of the unit circle has a range of values. , Represents pixels Hue components in the HSV color space The projection component along the vertical axis of the unit circle takes values ​​ranging from... cosine value Sine value The joint description of the hue's directional position on the unit circle constitutes the annular hue enhancement feature. Represents pixels The saturation component after normalization, i.e., the normalized saturation characteristic. Represents pixels The normalized luminance component, i.e., the normalized luminance feature; Represents pixels The normalized perceived brightness component, i.e., the normalized perceived brightness feature; Represents pixels The normalized red-green color components, i.e., the normalized red-green color characteristics; Represents pixels The normalized yellow-blue color components, i.e., the normalized yellow-blue color characteristics.

[0041] Using the current pixel and its eight neighboring pixels as the local analysis range, the dual color space consistency difference index is calculated. The dual color space consistency difference index is not simply the superposition of the color differences between the two color spaces, but a joint expression through geometric consistency coupling. The core idea is that only when both color spaces detect obvious color changes at the same time is the current pixel considered to be possibly at the boundary of a real object. If only one color space shows abnormal changes while the other color space remains stable, the abnormal changes are more likely to come from lighting disturbances, shadow effects, local noise, or the instability of the color space itself. In this case, the dual color space consistency difference index will not increase significantly.

[0042] The formula for calculating the consistency difference index between two color spaces is:

[0043] in, Represents pixels The consistency difference index of the two color spaces; Represents pixels The set of eight neighboring pixels; Indicates the number of neighboring pixels; Represents pixels The neighboring pixel positions; Indicates Euclidean distance; Represents pixels The HSV feature vector with ring-shaped hue enhancement. ; Represents pixels The HSV feature vector with ring-shaped hue enhancement; Represents pixels Normalized Lab-sensory color feature vector; Represents pixels Normalized Lab-sensory color feature vector; This indicates the characteristic differences in the HSV color space; This represents the characteristic differences in the Lab color space; This means that the feature differences between the HSV color space and the Lab color space are coupled through a product, ensuring that the product only increases when both color spaces show significant differences. If the difference between one color space is small, the overall difference is suppressed.

[0044] Histogram analysis was used to determine the consistency difference index between two color spaces. The distribution range of the difference index is traversed step by step according to the gray level division order of the difference index histogram between the minimum and maximum difference index to construct a candidate threshold set. The optimal automatic segmentation threshold is then selected from the candidate threshold set using the maximum inter-class variance method.

[0045] in, This represents the optimal automatic segmentation threshold; Indicates the candidate threshold; This indicates the selection of candidate thresholds that maximize the objective function within the brackets. This is used to select the threshold that maximizes the separation between low-difference pixel groups and high-difference pixel groups from all candidate thresholds. Low-difference pixel groups are defined by a dual color space consistency difference index less than or equal to... The pixels are composed of high-difference pixel groups, which are composed of pixels with a consistency difference index greater than 1 in the dual color space. The pixels that make up the whole; , These represent the ratios of the number of pixels in the low-difference pixel group and the high-difference pixel group to the total number of pixels in the remote sensing image to be processed, respectively. , These represent the mean values ​​of the dual color space consistency difference index for pixels in the low-difference pixel group and the high-difference pixel group, respectively. The global mean of the color space consistency difference index of a pixel.

[0046] A stable core pixel discrimination map is generated based on the optimal automatic segmentation threshold, and the stable core pixel discrimination result is obtained. The formula is expressed as follows:

[0047] in, This indicates the stable core pixel discrimination result. , representing pixels It belongs to the core area of ​​land features with stable color structure. , representing pixels It belongs to the boundary pixel, transition pixel, or mixed pixel;

[0048] To satisfy The pixels are searched for spatial connectivity using the 8-neighbor connectivity rule to obtain a stable set of candidate ground feature regions. :

[0049] in, Indicates the first A stable candidate feature area; Indicates the number of stable candidate feature regions;

[0050] Simultaneously define the set of boundary pixels to be backfilled. :

[0051] S2. Construct the regional color dominant feature vector of stable candidate land cover area through regional statistical method, and calculate the color matching difference between stable candidate land cover area and all land cover color prototypes in combination with land cover color prototypes to determine the land cover category of stable candidate land cover area; define neighboring stable candidate land cover areas, calculate the color belonging difference between the boundary pixel to be backfilled and the neighboring stable candidate land cover area, and determine the land cover category of the boundary pixel to be backfilled.

[0052] For each stable candidate feature region, the regional mean of the annular hue enhancement feature, normalized saturation feature, normalized brightness feature, normalized perceived brightness feature, normalized red-green contrast feature, and normalized yellow-blue contrast feature of all pixels in the region is calculated through regional statistical methods. The regional mean of the above features is combined to generate the regional color dominant feature vector, which can describe the color attributes of the feature region from the perspectives of color direction, color purity, brightness and darkness, and perceived color difference.

[0053] The specific representation of the region color-dominant feature vector is as follows:

[0054] in, Indicates the first The region color-dominant feature vector of a stable candidate land cover region; Indicates the first The average value of the hue cosine component of all pixels within the stable candidate feature region is used to characterize the . The average position of the color tone of a stable candidate feature area along the horizontal axis of a unit circle. The range of values ​​is ; Indicates the first The average value of the hue sinusoidal components of all pixels within the stable candidate feature region is used to characterize the ... The average position of the color tone of a stable candidate feature area along the vertical axis of a unit circle. The range of values ​​is ; Indicates the first The average normalized saturation of all pixels within a stable candidate feature region ; Indicates the first The average normalized brightness of all pixels within a stable candidate ground feature region. ; Indicates the first The average normalized perceived brightness of all pixels within a stable candidate ground feature region. ; Indicates the first The average value of the normalized red-green contrasting color components of all pixels within a stable candidate feature region. ; Indicates the first The average value of the normalized yellow-blue contrasting color components of all pixels within a stable candidate feature region. .

[0055] Furthermore, a prototype matching method based on color semantic prior is adopted to pre-establish a ground feature color prototype library. The ground feature color prototype library is constructed through statistical analysis of historical remote sensing sample data. Specifically, typical ground feature category sample areas such as vegetation, water bodies, bare land, buildings or hardened surfaces, shadows, and cloud, snow, or strong reflection areas are extracted from remote sensing image samples that have been manually annotated. For each ground feature category, the sample areas are converted to HSV color space and Lab color space, respectively, and the ring hue enhancement feature, normalized saturation feature, normalized brightness feature, normalized perceived brightness feature, normalized red-green contrast color feature, and normalized yellow-blue contrast color feature are extracted according to the color feature expression method in step S1, and a sample color feature vector is constructed. The sample color feature vectors of all sample areas of the same ground feature category are statistically averaged to obtain the standard color prototype vector corresponding to the ground feature category, which includes color attributes such as hue, saturation, brightness, perceived brightness, and contrast color, and is used to establish the correspondence between color features and ground feature semantic categories.

[0056] The color matching difference between each stable candidate land cover region and all land cover color prototypes is calculated using Euclidean distance to reflect the degree of closeness between the dominant color features of the region and the color prototypes of various land cover types. The smaller the color matching difference, the closer the color attributes of the current stable candidate land cover region are to the corresponding land cover color prototype. The minimum difference principle is adopted to determine the optimal land cover color prototype for each stable candidate land cover region, that is, the land cover color prototype with the smallest color matching difference is taken as the land cover category of the stable candidate land cover region.

[0057] For each pixel in the set of pixels to be backfilled, extract the ring-shaped hue-enhanced HSV feature vector and the normalized Lab-aware color feature vector, and construct the color feature vector of the pixel to be backfilled. To avoid the pixel to be backfilled being assigned to a spatially distant but color-similar region, the attribution determination is limited to the region of nearby stable candidate features. The expression for the set of nearby stable candidate feature regions is as follows:

[0058] in, Indicates the boundary pixels to be backfilled The set of neighboring stable candidate feature regions; Indicated by Centered on, with radius as The local neighborhood, when When, corresponding Neighborhood, when When, corresponding The neighborhood, as a specific embodiment, can be set to 1 here; This represents the empty set.

[0059] when When the search is an empty set, the neighborhood radius is gradually expanded until at least one stable candidate feature area is found. If no stable candidate feature area is found when the search radius is expanded to the maximum search radius, the pixel is marked as a pixel to be manually verified or an undetermined pixel. The maximum search radius is set to three times the search radius.

[0060] Furthermore, the Euclidean distance between the color feature vector of the boundary pixel to be backfilled and the regional color dominant feature vector of the neighboring stable candidate land cover area is calculated as the color attribution difference, which is used to measure the proximity between the boundary pixel to be backfilled and the neighboring stable candidate land cover area. The smaller the color attribution difference, the higher the color consistency and land cover continuity between the boundary pixel to be backfilled and the neighboring stable candidate land cover area.

[0061] Assigning the boundary pixels to be backfilled to the nearest stable candidate land cover region with the smallest color difference and inheriting the land cover category of the nearest stable candidate land cover region can effectively alleviate the boundary pixel labeling problem in complex scenes such as water-land junctions, building edges, and transition areas between vegetation and bare land, thereby improving the boundary holes, unlabeled areas, and regional breaks that occur in traditional hard threshold segmentation methods.

[0062] In summary, a remote sensing image classification and annotation method based on color space transformation has been completed.

[0063] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments; the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results; in some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0064] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0065] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A remote sensing image classification and annotation method based on color space conversion, characterized in that, Specifically, the following steps are included: S1. Acquire the remote sensing image to be processed, convert the image from RGB color space to HSV and Lab color spaces, and perform normalization processing to construct a ring-shaped hue-enhanced HSV feature vector and a normalized Lab perceptual color feature vector; based on the ring-shaped hue-enhanced HSV feature vector and the normalized Lab perceptual color feature vector, calculate the dual color space consistency difference index, and use the maximum inter-class variance method to obtain the optimal automatic segmentation threshold; based on the optimal automatic segmentation threshold, generate a stable core pixel discrimination map, and construct a stable candidate land cover region set and a boundary pixel set to be backfilled. S2. Construct the regional color dominant feature vector of stable candidate land cover area through regional statistical method, and calculate the color matching difference between stable candidate land cover area and all land cover color prototypes in combination with land cover color prototypes to determine the land cover category of stable candidate land cover area; define neighboring stable candidate land cover areas, calculate the color belonging difference between the boundary pixel to be backfilled and the neighboring stable candidate land cover area, and determine the land cover category of the boundary pixel to be backfilled.

2. The remote sensing image classification and annotation method based on color space conversion according to claim 1, characterized in that, S1 specifically includes: We introduce the HSV color space and the Lab color space, and extract the hue component, saturation component and lightness component of the HSV color space, as well as the perceived lightness component, red-green complementary color component and yellow-blue complementary color component of the Lab color space.

3. The remote sensing image classification and annotation method based on color space conversion according to claim 2, characterized in that, S1 specifically includes: Linear normalization is performed on the saturation component, brightness component, perceived brightness component, red-green contrasting color component, and yellow-blue contrasting color component to obtain normalized saturation features, normalized brightness features, normalized perceived brightness features, normalized red-green contrasting color features, and normalized yellow-blue contrasting color features. The hue component is converted into a circular vector representation to generate a circular hue enhancement feature. Based on the circular hue enhancement feature, normalized saturation feature, and normalized brightness feature, a circular hue enhancement HSV feature vector is constructed. Based on the normalized perceived brightness feature, normalized red-green contrasting color feature, and normalized yellow-blue contrasting color feature, a normalized Lab perceived color feature vector is constructed.

4. The remote sensing image classification and annotation method based on color space conversion according to claim 3, characterized in that, S1 specifically includes: Based on the HSV feature vector enhanced by ring tone and the normalized Lab perceptual color feature vector, using the current pixel and its eight neighboring pixels as the local analysis range, the feature differences between the current pixel and its neighboring pixels in the HSV color space and in the Lab color space are calculated. The feature differences in the HSV color space and the Lab color space are coupled to calculate the dual-color space consistency difference index. The specific calculation formula is as follows: in, Represents pixels The consistency difference index of the two color spaces; Represents pixels The set of eight neighboring pixels; Indicates the number of neighboring pixels; Represents pixels The neighboring pixel positions; Indicates Euclidean distance; Represents pixels The HSV feature vector with ring-shaped hue enhancement; Represents pixels The HSV feature vector with ring-shaped hue enhancement; Represents pixels Normalized Lab-sensory color feature vector; Represents pixels Normalized Lab-sensory color feature vector; This indicates the characteristic differences in the HSV color space; This indicates the characteristic differences in the Lab color space.

5. The remote sensing image classification and annotation method based on color space conversion according to claim 4, characterized in that, S1 specifically includes: The distribution range of the consistency difference index in the dual color space is statistically analyzed using histograms. A candidate threshold set is constructed by iterating through the minimum and maximum difference indices. Based on the consistency difference index and candidate thresholds, low-difference pixel groups and high-difference pixel groups are constructed. The maximum inter-class variance method is used to select the threshold that maximizes the separation between the low-difference and high-difference pixel groups as the optimal automatic segmentation threshold. Based on the optimal automatic segmentation threshold, a stable core pixel discrimination map is generated. Spatial connectivity search is performed on pixels belonging to the core regions of land features with stable color structures to construct a stable candidate land feature region set. Pixels not belonging to the core regions of land features with stable color structures are designated as boundary pixels to be backfilled, and a boundary pixel set to be backfilled is constructed.

6. The remote sensing image classification and annotation method based on color space conversion according to claim 1, characterized in that, S2 specifically includes: Typical land cover category sample areas are extracted from the manually labeled remote sensing image samples. For each land cover category, the sample areas are converted to HSV color space and Lab color space respectively, and the ring hue enhancement feature, normalized saturation feature, normalized brightness feature, normalized perceived brightness feature, normalized red-green contrast color feature and normalized yellow-blue contrast color feature are extracted according to the color feature expression method in step S1, and a sample color feature vector is constructed. The sample color feature vectors of all sample areas of the same land cover category are statistically averaged to obtain the standard color prototype vector corresponding to the land cover category.

7. The remote sensing image classification and annotation method based on color space conversion according to claim 6, characterized in that, S2 specifically includes: Based on the regional color dominant feature vector of stable candidate land cover areas and the standard color prototype vectors corresponding to all land cover categories, the color matching difference is calculated, and the land cover category of stable candidate land cover areas is determined by combining the principle of minimum difference.

8. The remote sensing image classification and annotation method based on color space conversion according to claim 7, characterized in that, S2 specifically includes: Based on the ring-shaped hue-enhanced HSV feature vector and the normalized Lab-aware color feature vector of the boundary pixels to be backfilled, a color feature vector of the boundary pixels to be backfilled is constructed. Based on the color feature vector of the boundary pixels to be backfilled and the regional color dominant feature vector of the neighboring stable candidate land cover area, the color assignment difference is calculated. The boundary pixels to be backfilled are assigned to the neighboring stable candidate land cover area with the smallest color assignment difference and inherit the land cover category of the neighboring stable candidate land cover area.

Images