Machine learning-based urban building color and style zoning control method and system

Abstract

The present application relates to the technical field of image recognition, in particular to a city building color and style zoning control method and system based on machine learning; the method comprises the following steps: acquiring a close-range street view image and a long-range view corridor image of a target view corridor, extracting a close-range building facade region, a long-range building facade region and a long-range natural background region by using an image semantic segmentation model, and determining corresponding building objects; calculating a close-range local color difference value and a long-range color contrast value of the building objects; merging the continuously distributed long-range building facade region into a long-range building group, and calculating a long-range contrast highlight intensity in combination with the long-range color contrast value and the horizontal projection continuity degree; calculating an inverted recognition value based on the close-range local color difference value and the long-range contrast highlight intensity, and dividing a long-range background contrast inversion control area and a conventional color coordination maintenance area; the present application can identify a building group with close-range color coordination but long-range background contrast highlight, and improve the accuracy of city building color and style zoning control.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, specifically to a method and system for controlling urban building color and appearance zoning based on machine learning. Background Technology

[0002] With the advancement of urban renewal, waterfront space development, mountain construction, and the management of important landscape corridors, the harmony between urban building colors and style has gradually become an important aspect of urban planning management, project review, and the improvement of urban spatial quality. For ordinary blocks, building color evaluation usually focuses on the main color, secondary color, brightness, saturation of the building facade, and its harmonious relationship with adjacent buildings. However, for style-sensitive areas such as mountainside, waterfront, urban gateways, and important visual corridors, the harmony of building colors depends not only on the building's own color or the local appearance in the near-view street interface, but also on the visual adaptation relationship between the building facade and the natural background such as mountains, water, sky, and vegetation in the distant view corridor. In waterfront areas, mountain-front construction interfaces, or open urban viewpoints, multiple buildings in the distant view can easily form continuous artificial color bands due to increased viewing distance and perspective compression. This causes light-colored, high-brightness, or high-contrast facades that are not obtrusive in the near-view street scene to appear as a prominent background contrast against the natural background in the distant view.

[0003] Existing methods for controlling the color and appearance of urban buildings largely rely on manual surveys, expert judgment, or building facade color recognition methods based on street view images. These methods can, to some extent, identify the main color of a building, differences in facade colors, and the local coordination between adjacent buildings. However, existing methods typically use the building facade colors in close-up street view images as the primary evaluation criterion, which easily equates the local coordination of a building in the close-up street interface with its overall coordination in the urban visual corridor. This makes it difficult to identify the problem of inverted evaluation where the same building or building group appears compliant in color in the close-up but jarring in the distant visual corridor due to the contrast of the natural background and the superposition of continuous horizontal projections. In particular, in mountain and waterfront visual corridors, the adjacency relationship between building facades and the natural background, the degree of continuous appearance of multiple building facades in distant images, and the inverse relationship between the local color differences in the close-up and the contrast of the distant background all directly affect the results of the style zoning control. Summary of the Invention

[0004] The purpose of this invention is to address the problem in the prior art where buildings along mountains, waterfronts, and important visual corridors appear harmonious in the foreground but jarring in the background due to natural background contrast and continuous artificial color bands, resulting in a reversed evaluation. The invention proposes a machine learning-based method and system for the zoning and control of urban building color and appearance.

[0005] The technical solution of this invention: a machine learning-based method for urban building color and style zoning control, comprising: S1. Obtain the near-view street scene image and the far-view corridor image of the target corridor. Use the image semantic segmentation model to extract the near-view building facade area, the far-view building facade area and the far-view natural background area, and determine the corresponding building objects between the near-view building facade area and the far-view building facade area. S2. Based on the representative colors extracted from the near-view building facade area and adjacent buildings, calculate the near-view local color difference value of the building object; S3. Extract representative colors based on the distant building facade area and the adjacent distant natural background area, and calculate the distant color contrast value of the building object; S4. Group the continuously distributed distant building facade areas into distant building groups, and calculate the distant contrast prominence intensity by combining the distant color contrast value and the horizontal projection continuity. S5. Based on the local color difference value of the foreground and the contrast intensity of the background of the corresponding building object, calculate the inverted recognition value of the background building group; S6. Based on the inverted recognition value, divide the distant background contrast inversion control area and the normal color coordination maintenance area.

[0006] Preferably, the foreground building facade area, the background building facade area, and the background natural background area are extracted, including: A trained image semantic segmentation model is used to perform image semantic segmentation on the near-view street scene image and the far-view corridor image respectively, and extract the corresponding near-view building facade area, far-view building facade area and far-view natural background area. The distant natural background area includes at least one of mountains, water bodies, sky, and vegetation.

[0007] Preferably, the same building that appears simultaneously in the close-up street view image and the distant view corridor image is identified as the corresponding building object.

[0008] Preferably, calculating the foreground local color difference value of the building object includes: Convert the close-up street scene image to the Lab color space; The median of the lightness and chromaticity components of the pixels within the foreground building facade area in the Lab color space is extracted and used as the representative color of the foreground building. Identify at least one adjacent building that is horizontally adjacent to the building object in the close-up street view image, and use the median of the close-up building representative color of the adjacent building as the close-up local reference color; Calculate the Euclidean distance between the representative color of the foreground building and the local reference color of the foreground to obtain the local color difference value of the foreground.

[0009] Preferably, calculating the background color contrast value of the building object includes: Convert the distant view corridor image to the Lab color space; The median of the pixels within the facade area of ​​the distant building is extracted in the Lab color space and used as the representative color of the distant building. Extract the area boundary of the distant building facade region; The boundary of the region is extended by adjacency according to the eight neighborhoods to obtain the boundary adjacency region surrounding the facade of the distant building. The intersection of the boundary adjacent region and the distant natural background region is performed to obtain the natural background pixels adjacent to the boundary of the distant building facade region. The median of the natural background pixels in the Lab color space is extracted as the representative color of the distant background; Calculate the Euclidean distance between the representative color of the distant building and the representative color of the distant background to obtain the color contrast value of the distant view.

[0010] Preferably, calculating the contrast intensity of the distant view includes: In the distant view corridor image, with the same continuous distant natural background area as a reference, distant building facade areas with contiguous boundaries and continuous horizontal projection in the lateral direction are grouped into the distant building group. The horizontal projection continuity is calculated based on the coordinate range of the distant building group in the horizontal direction of the distant view corridor image and the distribution of building pixels. The overall color contrast of the building group is obtained by weighting the visible pixel area of ​​the building object in the distant view corridor image and calculating the distant color contrast value of the building object. The product of the horizontal projection continuity and the overall color contrast of the building group is used as the intensity of the background contrast.

[0011] Preferably, calculating the inverted identification value of the distant building group includes: The foreground local color difference of the building object is obtained by weighting the visible pixel area of ​​the building object in the foreground street view image and calculating the foreground local color difference of the building object. Calculate the median and median absolute deviation of the distant contrast prominence intensity sequence for all distant building groups, and the median and median absolute deviation of the foreground local color difference sequence. Using the median and the absolute deviation of the median, adaptive normalization processing is performed on the contrast intensity of distant buildings and the local color difference of foreground buildings in each distant building group. The inverted recognition value is obtained by subtracting the normalized local color difference of the foreground from the normalized contrast intensity of the distant view.

[0012] Preferably, the distant background contrast inversion control area and the normal color coordination preservation area are divided based on the inverted recognition value, including: Construct an inverted identification value sequence for all the aforementioned vision building groups, and extract the median of the inverted identification value sequence as the adaptive inverted baseline value; The distant building group whose inverted identification value is greater than zero and greater than the adaptive inverted reference value is identified as the distant background contrast inverted building group. Grouping the spatially adjacent groups of inverted background buildings into a background contrast control zone, and grouping the remaining groups of background buildings into a regular color coordination maintenance zone.

[0013] Preferably, the same building that appears simultaneously in both the close-up street view image and the distant view corridor image is identified as the corresponding building object, including: Acquire at least one of the following: shooting location information, shooting direction information, and building area geometric information corresponding to the near-view street scene image and the far-view corridor image; Based on at least one of the shooting location information, shooting direction information, and building area geometric information, a candidate correspondence between the close-up building facade area in the close-up street view image and the distant building facade area in the distant view corridor image is determined. Extract the center position of the outline, the horizontal width ratio, and the vertical height ratio of the near-view building facade area and the far-view building facade area; Based on the candidate correspondence, the center position of the outline, the horizontal width ratio, and the vertical height ratio, the same building object corresponding to the near-view building facade area and the far-view building facade area is determined.

[0014] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: This invention simultaneously acquires near-field street view images and far-field corridor images of a target visual corridor, and uses an image semantic segmentation model to extract the near-field building facade area, the far-field building facade area, and the far-field natural background area. This expands the evaluation of building color from a single near-field street interface to a visual corridor analysis scale combining near and far views. By calculating the near-field local color difference value of a building relative to adjacent buildings, it can reflect the degree of local harmony of the building under close-up street observation. By calculating the far-field color contrast value of a building relative to the far-field natural background, it can reflect the visual contrast relationship between the building and the background such as mountains, water bodies, sky, and vegetation in mountainous, waterfront, or important visual corridors. It can distinguish and evaluate the harmony of adjacent buildings in the near-field and its adaptability to the natural background in the far-field, avoiding judging the harmony of building style solely based on the near-field street view color results, thereby improving the accuracy and scene adaptability of urban building color and style recognition. This invention further groups continuously distributed distant building facade areas into distant building groups, and calculates the intensity of distant contrast prominence by combining distant color contrast values ​​and horizontal projection continuity. This allows the visual magnification effect of multiple building facades in the distance being compressed by perspective to form continuous artificial color bands to be quantified. By calculating inverted identification values ​​based on local color difference values ​​in the foreground and the intensity of distant contrast prominence, the invention can identify the evaluation inversion phenomenon where building groups exhibit relatively harmonious colors in the foreground but form a significant prominence contrast with the natural background in the distance. Based on the inverted identification values, the invention divides the distant background contrast inversion control area and the conventional color harmony maintenance area, so that the zoning results can directly correspond to key control objects along mountains, waterfronts, and important visual corridors. This provides a quantifiable and verifiable zoning basis for urban building color review, waterfront interface style control, mountain front construction interface management, and important landscape corridor style optimization. Attached Figure Description

[0015] Figure 1 This is a flowchart of the urban building color and style zoning control method based on machine learning proposed in this invention. Figure 2 This is a block diagram of the urban building color and style zoning control system based on machine learning proposed in this invention. Detailed Implementation

[0016] Example 1, as Figure 1 As shown, the urban building color and style zoning control method based on machine learning proposed in this invention includes: This embodiment provides a machine learning-based method for urban building color and style zoning control, applicable to mountain corridors, waterfront corridors, important urban landscape corridors, urban gateway landscape interfaces, and urban spaces where the compatibility between building colors and natural backgrounds needs to be evaluated. It is particularly suitable for the following situations: buildings appear similar in color to adjacent buildings in close-up street view images, showing local color harmony in the foreground; however, in distant view corridor images, multiple building facades appear together with natural backgrounds such as mountains, water bodies, sky, or vegetation, and the building facades form continuous artificial color bands in the distance, resulting in a significant contrast between the building group and the natural background. This embodiment identifies this phenomenon of inverted foreground color compliance and prominent contrast with the distant background, generating a distant background contrast inversion control zone and a normal color harmony maintenance zone. It should be noted that, in this embodiment, the target visual corridor refers to the urban space range from which the building facade area and the natural background area together form a visual interface, which can be observed from a predetermined observation position; the predetermined observation position includes waterfront walkways, bridges, opposite banks, open spaces in front of mountains, urban viewing nodes, road nodes, or important viewpoints determined by planning; the target visual corridor is not limited to a single linear passage, but includes the visual analysis range formed by the observation position, observation direction, visible building group, and the natural background area that images together with the building group; Specifically, close-up street view images refer to images collected from close-up observation positions of streets, sidewalks, street-facing spaces, or buildings, used to reflect the color relationship between a building and its lateral adjacent buildings in the close-up street interface; distant view corridor images refer to images collected from predetermined observation positions in the target corridor, used to reflect the color relationship between a building or building group and natural backgrounds such as mountains, water bodies, sky, and vegetation appearing together in the distant view field; close-up street view images are used to calculate the local color difference of a building relative to its adjacent buildings in the close-up view, while distant view corridor images are used to calculate the color contrast of a building relative to its natural background in the distance. The two types of images correspond to color evaluation criteria at different observation scales; The machine learning model in this embodiment includes at least an image semantic segmentation model. It should be noted that the image semantic segmentation model in this embodiment refers to a machine learning model capable of outputting a corresponding category label for each pixel in the input image. The category labels include at least building facade category, natural background category, and non-target category. For close-up street scene images, the image semantic segmentation model outputs at least building facade category and non-target category. For distant view corridor images, the image semantic segmentation model outputs at least building facade category, mountain category, water category, sky category, vegetation category, and non-target category. The output of the image semantic segmentation model is a pixel category matrix with the same size as the input image. This pixel category matrix is ​​used to generate the building facade area and distant natural background area required for subsequent calculations. Specifically, the training samples for the image semantic segmentation model include near-view street scene image samples and far-view corridor image samples. The annotation results of the near-view street scene image samples include building facade pixel annotations and non-target pixel annotations. The annotation results of the far-view corridor image samples include building facade pixel annotations, mountain pixel annotations, water pixel annotations, sky pixel annotations, vegetation pixel annotations, and non-target pixel annotations. During training, the image samples are input into the image semantic segmentation model, and the corresponding pixel category annotation map is used as the supervision label. The model parameters are trained through cross-entropy loss, Dice loss, or a combination of cross-entropy loss and Dice loss, so that the pixel category matrix output by the model can distinguish between building facade areas and far-view natural background areas. Specifically, the image semantic segmentation model can be U-Net, DeepLab, SegFormer, Mask2Former, or other semantic segmentation models that can output pixel-level category matrices; the specific network structure of the model is not a limitation of this embodiment, as long as it can output building facade category pixels and natural background category pixels based on the input near-field street scene image and far-field corridor image, it can be used to implement the region extraction process of this embodiment. Specifically, in the model inference stage, the near-view street scene image is input into the trained image semantic segmentation model to obtain a near-view pixel category matrix, and building facade category pixels are extracted from the near-view pixel category matrix. Connectivity analysis is performed on the building facade category pixels to determine each connected building facade region as a near-view building facade region. The distant view corridor image is input into the trained image semantic segmentation model to obtain a distant view pixel category matrix, and building facade category pixels, as well as mountain category, water category, sky category, and vegetation category pixels are extracted from the distant view pixel category matrix. Connectivity analysis is performed on the distant view building facade category pixels to determine each connected building facade region as a distant view building facade region. The mountain category, water category, sky category, and vegetation category pixels are merged to obtain the distant natural background region. Thus, the input of the image semantic segmentation model is image data, and the output is a set of pixel regions that can be directly used for subsequent color difference calculation and region adjacency calculation.

[0017] The method in this embodiment includes the following steps; S1. Obtain the near-view street scene image and the far-view corridor image of the target corridor, and determine the corresponding building objects in the near and far views; Acquire near-view street scene images and far-view street scene images of the target visual corridor; the near-view street scene image is denoted as... The distant view corridor image is recorded as In this context, the superscript N indicates a close-up view, and the superscript F indicates a distant view; close-up street view image. Used to reflect the color harmony of a building relative to adjacent buildings when viewed up close on a street, a distant view corridor image. Used to reflect the color contrast of architectural objects relative to the natural background in a distant view corridor; Before performing image semantic segmentation and color calculation, the close-up street scene image... and distant view gallery images Image preprocessing is performed; the image preprocessing includes at least one of white balance correction, brightness normalization and abnormal pixel removal; white balance correction is used to reduce the overall color cast caused by different shooting devices or different shooting times; brightness normalization is used to reduce the impact of exposure differences on the brightness component; abnormal pixel removal is used to remove abnormal pixels in overexposed areas, strong reflective areas, deep shadow areas and obviously occluded areas. Specifically, abnormal pixel removal can be judged based on pixel luminance components and luminance distribution within the building facade area. When the luminance component of a pixel within the building facade area deviates significantly from the median and luminance dispersion range of the building facade area, the pixel is considered an abnormal pixel and is not included in the median statistics when calculating the representative color. It should be noted that in this embodiment, the near-view local color difference value is calculated within the same near-view street scene image, and the far-view color contrast value is calculated within the same far-view corridor image. The absolute color values ​​between the near-view street scene image and the far-view corridor image are not directly compared. The above image preprocessing is used to improve the stability of color calculation within the same image. Collecting close-up street view images At the same time, the image should include the visible facade area of ​​the target building and its horizontally adjacent buildings; when acquiring distant view corridor images... When the image contains the target building group and the natural background area adjacent to or jointly imaged with the building group, the natural background area includes at least one of mountains, water bodies, sky, and vegetation. For waterfront promenades, the distant natural background area typically includes water bodies, sky, and waterfront vegetation. For mountain promenades, the distant natural background area typically includes mountains, sky, and mountain vegetation. For important urban landscape corridors, the distant natural background area may include sky, green space, water surface, or mountain background. Near-view street view image Inputting a trained image semantic segmentation model yields the foreground building facade region; the background corridor image is then processed. Inputting a trained image semantic segmentation model yields the distant building facade region and the distant natural background region; for the i-th building object, its near-field building facade region is denoted as... Its distant building facade area is recorded as The distant natural background area is denoted as ; It should be noted that the foreground building facade region refers to the set of pixels in the foreground street view image where the i-th building object is identified as the building facade, and is used to represent the visible color of the building object in the foreground street interface; the background building facade region refers to the set of pixels in the background view image where the i-th building object is identified as the building facade, and is used to represent the visible color of the building object in the background view. Both the foreground and background building facade regions are pixel regions output by the image semantic segmentation model and obtained through connected component analysis. The two are associated with the same building entity through the correspondence between building objects. It should be noted that the distant natural background area refers to the set of natural element pixels in the distant view corridor image that together with the building facade constitute the distant visual background; the natural elements include at least one of mountains, water bodies, sky, and vegetation; the distant natural background area does not include roads, vehicles, bridge railings, billboards, temporary structures, or other non-natural background areas; the distant natural background area is used to determine the natural background pixels that are visually adjacent to the building facade area, and serves as a background reference in the calculation of the distant color contrast value; This represents the i-th building object in the close-up street view image. The set of building facade pixels in the text. This represents the i-th building object in the distant view corridor image. The set of building facade pixels in the text. Image representing a distant view corridor A collection of natural background pixels.

[0018] For the merged distant natural background area Connectivity analysis is performed to obtain at least one connected region of distant natural background. Each connected region of distant natural background represents a set of spatially continuous natural background pixels in the distant view corridor image. If multiple natural background categories, such as mountains, water bodies, sky, and vegetation, exist simultaneously in the same distant view corridor image, adjacent or contiguous natural background category pixels are first merged into a continuous natural background region. Then, connected component analysis is used to determine each continuous natural background region. When merging distant building groups, the same continuous natural background region is used as a reference to ensure that the division of distant building groups corresponds to the natural background that the buildings are actually adjacent to in the view corridor. Will appear simultaneously in the close-up street view image and distant view gallery images The same building in the list is identified as the corresponding building object, resulting in a collection of building objects: ; in, Let i represent the set of building objects, i represent the i-th building object, and n represent the total number of building objects in the target visual corridor that have been identified and whose near and far views have been completed; It should be noted that a building object refers to an analytical object that can be identified in both near-view street scene images and far-view corridor images, and can be determined as the same building entity through shooting location, shooting direction, geometric information of building area, building number, building outline vector data, or 3D city model projection results; a building object is not simply a connected region of the image, but the same building entity after the correspondence between the near-view building facade area and the far-view building facade area is completed; subsequent near-view local color difference value, far-view color contrast value, and inversion recognition value are all based on the building object or the far-view building group composed of building objects; Specifically, the correspondence between the near-view building facade area and the far-view building facade area is determined through candidate matching and matching cost calculation; at least one of the following is obtained: shooting location information, shooting direction information, and building area geometric information corresponding to the near-view street view image and the far-view corridor image; based on the shooting location information and shooting direction information, the building facade areas in the near-view street view image that may appear in the far-view corridor image are determined, forming a candidate set of near-view building facade areas; based on the far-view building facade areas in the far-view corridor image, a candidate set of far-view building facade areas is formed. For any candidate near-view building facade area and any candidate building facade area Extract the center position of its outline, the horizontal width ratio, and the vertical height ratio; for the near-view building facade area. The center position of the outline is denoted as The horizontal width ratio is denoted as The vertical height ratio is denoted as ; Visible building facade area The center position of the outline is denoted as The horizontal width ratio is denoted as The vertical height ratio is denoted as Where u represents the sequence number of the candidate near-view building facade area, v represents the sequence number of the candidate far-view building facade area, the center position of the outline is the normalized coordinate of the center point of the outer rectangle of the building facade area in the corresponding image coordinate system, the horizontal width ratio is the ratio of the width of the outer rectangle of the building facade area to the width of the corresponding image, and the vertical height ratio is the ratio of the height of the outer rectangle of the building facade area to the height of the corresponding image. Calculate candidate near-view building facade areas With candidate building facade area Matching cost: ; in, Indicates the candidate near-view building facade area With candidate building facade area The matching cost; Indicates the difference in the position of the center of the contour; Indicates the difference in horizontal width ratio; Indicates the difference in vertical height ratio; This represents the geometric candidate deviation determined by the shooting position information and shooting direction information; if shooting position information or shooting direction information is not available, then let... The value is 0, and matching is performed based on the geometric information of the building area; Based on matching cost The method employs a minimum cost matching approach to determine the one-to-one correspondence between near-view and far-view building facade areas. Specifically, for each far-view building facade area, the near-view building facade area with the lowest matching cost is selected from the candidate near-view building facade areas as the corresponding object. When multiple far-view building facade areas correspond to the same near-view building facade area, the correspondence with the lowest matching cost is retained, and the remaining far-view building facade areas are re-matched with the remaining candidate near-view building facade areas. This process identifies the same building that appears simultaneously in both the near-view street view image and the far-view corridor image, and treats it as the building object. It should be noted that when building numbers, building outline vector data, planning plot data, or 3D city models exist, the consistency of building numbers, the overlap of building outline projections, or the projection results of 3D city models can be used as constraints on candidate correspondences to reduce mismatches caused by differences in near and far perspectives.

[0019] S2. Calculate the local color difference value of the building object in the foreground based on the foreground building facade area and adjacent buildings; Near-view street view image Convert to Lab color space; for close-up street scene images For any pixel p in the color space, its color vector is denoted as: ; in, This represents the color vector of pixel p in a close-up street view image. This represents the luminance component of pixel p. and This represents the chromaticity component of pixel p; For the i-th building object, in its foreground building facade area Extracting representative colors from nearby buildings: ; in, This represents the foreground architectural color of the i-th building object. This indicates the number of pixels within the region. Quantity, Components and The median of the components is used. It should be noted that the representative color refers to the color vector used to characterize the overall visible color of a specified pixel area. In this embodiment, the representative color is composed of the median of the lightness and chromaticity components of each pixel in the Lab color space. Using the median to extract the representative color can reduce the impact of local shadows, strong reflections, occlusions, stains, and image noise on the overall color of the area. The representative color of the foreground buildings is used to characterize the building facade color of the building object in the foreground street view image, the representative color of the background buildings is used to characterize the building facade color of the building object in the background corridor image, and the representative color of the background is used to characterize the natural background color adjacent to the background building facade area. Representative colors of near-view architecture Expanded to: ; in, This represents the lightness component of the foreground building color of the i-th building object. and This represents the chromaticity component of the foreground building color of the i-th building object.

[0020] Determine the i-th building object in the close-up street view image At least one adjacent building in the horizontal direction forms a set of adjacent building objects: ; in, Let j represent the set of adjacent building objects of the i-th building object, and j represent adjacent building objects. It should be noted that adjacent buildings refer to buildings that are located on the same street surface as the current building object in the close-up street view image, and are located to the left or right of the current building object in the horizontal direction of the image. If there are identifiable buildings on both sides of the current building object, then both buildings can be considered adjacent buildings. If there is only one identifiable building on the current building object, then the building on that side is considered an adjacent building. If the current building object does not have any directly horizontally adjacent buildings in the current close-up street view image, then the building object with the closest horizontal distance to the current building object in the same street surface is considered an adjacent building. Adjacent buildings are used to form local reference colors in the close-up view, thereby evaluating the local color differences of the current building object in the close-up street surface. Calculate the local reference color of the i-th building object based on the representative colors of the foreground buildings of each adjacent building object in the set N(i): ; in, This represents the foreground local reference color of the i-th building object. The color representing the foreground building of the j-th adjacent building object; Near-view reference color Expanded to: ; in, This indicates the lightness component of the reference color in a foreground area. and This indicates the chromaticity components of the reference color in a foreground area; Calculate the foreground local color difference value of the i-th building object: ; in, This indicates the degree of local color difference between the i-th building object and its adjacent buildings in the foreground street view; The smaller the value, the closer the color of the building object is to that of adjacent buildings in the close-up street view image, and the more harmonious the local colors in the foreground. The larger the value, the more obvious the color difference between the building and its neighboring buildings in the foreground street view image, and the higher the degree of abruptness in the foreground. The basis of this calculation is that the color vector distance in the Lab color space can be used to characterize the color difference in visual perception. For urban building color control, neighboring buildings in the foreground street interface usually constitute direct visual references. Therefore, the Euclidean distance between the building's representative color and the reference color of the neighboring buildings can characterize the local color coordination relationship of the building at the foreground street scale.

[0021] S3. Calculate the color contrast value of the building object in the distance based on the building facade area and the adjacent distant natural background area; View corridor image Convert to Lab color space; for distant gallery images For any pixel q, its color vector in the Lab color space is denoted as: ; in, This represents the color vector of pixel q in the distant view corridor image. This represents the luminance component of pixel q. and Represents the chromaticity component of pixel q; For the i-th building object, in its distant building facade area Extracting the representative colors of distant buildings: ; in, Represent the background building color of the i-th building object; use the background building color... Expanded to: ; in, This represents the lightness component of the background building color of the i-th building object. and This represents the chromaticity component of the background building color of the i-th building object; Determine the distant building facade area relative to the i-th building object. Natural background pixels adjacent to the boundary; specifically, extracting the facade area of ​​distant buildings. The boundary of the region is denoted as: ;in, This represents the boundary of the distant building facade area of ​​the i-th building object; According to the eight neighboring areas, the regional boundaries are defined. By performing adjacency expansion, the area surrounding the building facade in the distance is obtained. The boundary adjacent region is denoted as: ,in, This represents the boundary adjacency region surrounding the distant building facade area of ​​the i-th building object; the eight-neighborhood refers to the eight adjacent pixel positions of a pixel, namely its top, bottom, left, right, top left, top right, bottom left, and bottom right; using the eight-neighborhood can identify horizontally, vertically, and diagonally adjacent natural background pixels, thereby extracting background pixels that are visually adjacent to the building outline more completely. It should be noted that the boundary adjacency region refers to the set of pixels obtained by extending outward from the boundary of the distant building facade area according to the pixel adjacency relationship. The boundary adjacency region is used to find natural background pixels that are visually adjacent to the distant building facade area. The boundary adjacency region is used instead of the entire natural background area in the entire distant view corridor image in order to make the representative color of the distant background reflect the natural background color that actually occurs in visual contact with the building object, thereby improving the relevance of the distant color contrast value.

[0022] Adjacent areas of the boundary With the distant natural background area Perform an intersection operation to obtain the set of natural background pixels adjacent to the boundary of the distant building facade area: ; in, Represents the set of distant natural background pixels adjacent to the i-th building object; when When the set is empty, it indicates that no natural background pixels adjacent to the boundary of the distant building facade area were obtained according to the first eight-neighbor adjacency expansion; in this case, along the distant building facade area The region boundary is extended layer by layer through adjacency; the r-th layer of the extended region is denoted as . where r is a positive integer. This refers to the boundary adjacency region obtained by performing an adjacency extension on the region boundary according to the eight-neighbor domain. In order to be in Based on this, continue to expand outwards according to the eight-neighbor domain to obtain the adjacent region; For each adjacent extended region, calculate: ; in, Let r represent the set of pixels in the adjacent extended region of the r-th layer that belong to the distant natural background region; if there exists a smallest positive integer r such that If not empty, then As the set of distant natural background pixels adjacent to the i-th building object, and based on Extract the representative color of the distant background; If, after extending the building object layer by layer to the outer boundary of the distant building group to which it belongs, no pixels belonging to the distant natural background area are obtained, then the distant color contrast value of the building object is not calculated separately. Instead, the building object is incorporated into its distant building group, and the distant color contrast of the building group is represented by other building objects in the distant building group that can obtain adjacent distant natural background pixels. If none of the building objects in the distant building group can obtain adjacent distant natural background pixels, then the boundary adjacency extension is performed based on the overall outer boundary of the distant building group, and the distant natural background pixels obtained after the extension are used as the background reference pixels of the distant building group.

[0023] Based on the set of adjacent distant natural background pixels Extract the representative color of the distant background: ; in, This represents the background color corresponding to the i-th building object; if The result is obtained by using the adjacency expansion of the r-th level when the space is empty. Then in the above formula Depend on Replacement; if the overall background reference pixels of the distant building group are used, then in the above formula... The background pixels of the entire distant architectural group were replaced; The representative color of the background Expanded to: ; in, This indicates the brightness component of the color representing the background color. and This indicates the chromaticity components of the color representing the background color. Calculate the background color contrast value of the i-th building object: ; in, This represents the degree of color contrast of the i-th building object relative to the adjacent natural background in the distant view corridor image; The larger the value, the more obvious the color difference between the building and the background of mountains, water, sky, or vegetation in the distance, and the easier it is to form a prominent contrast in the distance view corridor; The smaller the value, the closer the building object is to the adjacent natural background color in the distance, and the weaker the background contrast. This calculation is based on the fact that buildings in scenes along mountains, waterfronts, or important viewpoints are not observed in isolation, but enter the observer's field of view together with the natural background. The greater the color difference between the building and its adjacent natural background, the easier it is for the building to be perceived as abrupt in the distance. By selecting natural background pixels adjacent to the boundary of the building facade area in the distance, rather than selecting all natural background pixels in the entire distance image, the representative color of the distance background can more accurately reflect the background environment that the building object actually encounters at the visual boundary of the distance.

[0024] S4. Group the continuously distributed facade areas of distant buildings into distant building groups and calculate the contrast intensity of the distant view. Specifically, in the distant view corridor image In this context, using the same continuous distant natural background area as a reference, distant building facade areas with contiguous boundaries and continuous horizontal projections in the lateral direction are grouped into distant building groups, resulting in a set of distant building groups: ; in, Indicates a group of buildings in the future. represents the g-th distant building group, and m represents the number of distant building groups within the target visual corridor; It should be noted that a distant building group refers to a collection of multiple building objects that are adjacent to the same continuous natural background area in a distant view corridor image and appear in a continuous or near-continuous relationship in the horizontal direction of the image; a distant building group is used to represent a continuous artificial color band formed by multiple building facades in a distant view corridor; a single building object can constitute a distant building group, and multiple building objects can also constitute the same distant building group due to continuous horizontal projection, continuous natural background adjacency, or being on the same visible corridor interface after partial occlusion; For the g-th vision building group, the set of building objects it contains is denoted as: ; in, This represents the set of building objects corresponding to the g-th vision building group; The purpose of dividing the distant building group is to identify whether multiple buildings form a continuous artificial color band in the distant view corridor; even if a single building object has a color difference with the natural background, it may not necessarily constitute an obvious risk to the view corridor appearance; when multiple building objects appear continuously in the horizontal direction in the distant image, the color difference of the building facade will be amplified in the form of a continuous interface, thus forming a phenomenon of prominent contrast of the distant background. Specifically, the distant building group is determined jointly based on the horizontal projection range and the adjacency relationship with the natural background; for any distant building facade area... Calculate its projection range in the horizontal direction of the distant view corridor image: ; in, This represents the projection range of the distant building facade area of ​​the i-th building object in the horizontal direction. Indicates the building facade area in the distance. The minimum x-coordinate of the building pixels in the middle. Indicates the building facade area in the distance. The maximum horizontal coordinate of the building pixels in the middle; For any two distant building facade areas and If both satisfy at least one of the following conditions, they are identified as mergeable distant building facade areas: First, the lateral projection ranges of the two... and First, they are connected or overlapped; second, they are both adjacent to the same continuous connected region of distant natural background, and there are no other continuous building interfaces blocking them in the horizontal direction of the distant view corridor image; third, the two are only separated by pixel intervals from each other by tree canopies, road facilities, bridge railings or non-buildings, but they together with the same continuous natural background region form the same visible corridor interface. Multiple distant building facade areas that satisfy the above-mentioned merging relationship and are continuously connected through the merging relationship are merged into the same distant building group; even if there is partial obstruction or small-scale non-building intervals between distant building facade areas, as long as they form a horizontal continuous display relationship in front of the same continuous natural background, they can still be merged into the same distant building group, thereby reflecting the actual visual experience of continuous artificial color bands in the distant view corridor.

[0025] For the g-th distant building group, calculate its coordinate range in the horizontal direction of the distant view corridor image; let the minimum horizontal coordinate of the facade area of ​​all distant buildings in this distant building group be . The maximum value of the x-coordinate is If there exists a vertical coordinate y on the horizontal coordinate x such that the pixel (x,y) belongs to the facade area of ​​any distant building in the distant building group, then it is considered that there is a building pixel in the horizontal coordinate column. Calculate the horizontal projection continuity of the g-th distant building group: ; in, This represents the horizontal projection continuity of the g-th distant building group. Let x represent the minimum x-coordinate of the g-th distant building group in the horizontal direction of the distant view corridor image. Let (x, y) represent the maximum x-coordinate of the g-th distant building group in the horizontal direction of the distant view corridor image, and (x, y) represent the pixel coordinates in the distant view corridor image. This indicates an indicator function; when there is a building pixel belonging to the g-th distant building group in the vertical column containing the x-coordinate, The value of is 1; otherwise, The value of is 0; It should be noted that horizontal projection continuity refers to the degree to which a group of distant buildings continuously occupies a column of images in the horizontal direction of the distant view corridor image. For the coordinate range of the distant building group in the horizontal direction, if most of the image columns within the coordinate range contain building pixels, it indicates that the distant building group appears as a continuous artificial color band in the distant view, and the horizontal projection continuity is high. If there are many image columns without building pixels within the coordinate range, it indicates that the distant building group appears as a discontinuous state in the distant view, and the horizontal projection continuity is low. Horizontal projection continuity is used to characterize whether the color contrast of the distant view is amplified by the continuous building interface.

[0026] By utilizing the visible pixel area of ​​building objects in the distant view corridor image, the color contrast value of the building object in the distance is weighted and calculated to obtain the overall color contrast of the g-th distant building group: ; in, This represents the overall color contrast of the g-th distant building group. This represents the visible pixel area of ​​the i-th building object in the distant view corridor image. This represents the background color contrast value of the i-th building object. The weighting is based on the visible pixel area of ​​the background because building objects with larger visible areas in the background image contribute more to the overall visual perception, while building objects with smaller visible areas or more obscuration contribute relatively less to the visual contrast of the background corridor.

[0027] Horizontal projection continuity Contrast with the overall color scheme of the building complex The product of these factors enhances the intensity of the contrast in the foreground. ; in, This indicates the intensity of the foreground contrast of the g-th foreground building group; The larger the value, the stronger the color contrast between the distant building group and the natural background, and the stronger the continuous display characteristic in the distant view corridor image, making it easier to create a prominent contrast with the distant background. This calculation is based on the fact that the abruptness of a distant view depends not only on the color difference between a single building and the natural background, but also on whether this difference is presented as a continuous interface in the view corridor; horizontal projection continuity. Used to characterize the degree to which a group of buildings forms a continuous artificial color band in a distant image, the overall color contrast of the building group. Used to characterize the color difference of the building group relative to the natural background, the product of the two can express the degree to which the contrast of the distant background is amplified by the continuous artificial color band; It should be noted that the intensity of the contrast in the distant view refers to the overall visual prominence of the architectural group relative to the natural background in the distant view. It takes into account both the overall color contrast of the architectural group and the continuity of the horizontal projection. The greater the overall color contrast of the architectural group, the more obvious the color difference between the architectural group and the natural background. The higher the continuity of the horizontal projection, the easier it is for the observer to perceive the color difference in the form of a continuous artificial color band. The intensity of the contrast in the distant view is used to characterize the degree of abruptness of the architectural group relative to the natural background in the distant view corridor.

[0028] S5. Calculate the inverted recognition value based on the local color difference value of the near view and the contrast intensity of the far view; Specifically, for the g-th vision building group, based on the set of building objects it contains... By utilizing the visible pixel area of ​​building objects in the foreground street view image, the local color difference values ​​of the building objects in the foreground are weighted and calculated to obtain the local color difference of the foreground corresponding to the distant building group: ; in, This represents the local color difference in the foreground corresponding to the g-th distant building group. This represents the visible pixel area of ​​the i-th building object in the close-up street view image. This represents the local color difference value in the foreground of the i-th building object; The weighting is based on the visible pixel area in the foreground because building objects with larger visible areas in the foreground image contribute more to the perception of foreground color in the street interface, while building objects with smaller visible areas or more obscuration contribute relatively less to the local color difference in the foreground. Construct a sequence of perspective contrast intensity for all perspective building groups: ; in, This represents the sequence of perspective contrast intensity for all perspective building groups. Construct a sequence of foreground local color differences for all distant building groups: ; in, This represents the sequence of local color differences in the foreground corresponding to all distant building groups. Calculate the contrast salience intensity sequence of the distant view The median of: ; in, The median of the sequence representing the intensity of contrast in the distant view; Calculate the contrast salience intensity sequence of the distant view absolute deviation of the median: ; in, This represents the median absolute deviation of the sequence indicating the intensity of the contrast in the distant view.

[0029] Calculate the local color difference sequence in the foreground The median of: ; in, This represents the median of the local color difference sequence in the foreground. Calculate the local color difference sequence in the foreground absolute deviation of the median: ; in, This represents the median absolute deviation of the local color difference sequence in the foreground. Adaptive normalization is applied to the perspective contrast intensity of the g-th perspective building group using the median and the absolute deviation of the median: ; in, This represents the normalized perspective contrast intensity of the g-th perspective building group. Indicates the calculation of the stable term; Using the median and the absolute deviation of the median, adaptive normalization is performed on the local color differences in the foreground corresponding to the g-th distant building group: ; in, This represents the normalized local color difference in the foreground corresponding to the g-th distant building group; Adaptive normalization using the median and median absolute deviation is intended to make the evaluation results comparable under different target view corridors, different natural backgrounds, and different urban color benchmarks, while reducing the impact of a small number of abnormal building groups on the normalization results.

[0030] The inverted recognition value of the g-th distant building group is obtained by subtracting the normalized local color difference of the foreground from the normalized contrast intensity of the distant view: ; in, The identification value represents the inversion of the foreground color compliance and the contrast between the background and the background in the g-th distant building group; It should be noted that the inversion recognition value is a numerical value used to characterize whether a distant building group exhibits a degree of inversion where the foreground colors are relatively harmonious while the distant background contrasts sharply. The inversion recognition value is obtained by subtracting the normalized local color difference of the foreground from the normalized contrast intensity of the distant view. When the color difference of the distant building group relative to adjacent buildings is low in the foreground street view image, but the contrast intensity relative to the natural background is high in the distant view corridor image, the inversion recognition value increases. The larger the inversion recognition value, the more the distant building group conforms to the inversion feature of seemingly harmonious foreground but prominent foreground background relative to natural background. When the color difference between the g-th distant building group and its adjacent buildings in the foreground street view image is low, Relatively low; when the distant building group has high color contrast with the natural background in the distant view corridor image and forms a continuous artificial color band. Relatively high; at this time, An increasing value indicates that the distant building group conforms to the inverted characteristic of relatively harmonious colors in the foreground and prominent contrast in the background; conversely, if the contrast in the background is not strong, or if the local color difference in the foreground is already high, then... It will not increase significantly, and this building group is not among the objects in which near-view compliance and far-view prominence are reversed, which are the focus of this embodiment.

[0031] S6. Divide the distant background contrast inversion control area and the normal color coordination maintenance area based on the inverted recognition value; Specifically, construct the inverted identification value sequence for all vision building groups: ; in, This represents the inverted sequence of identification values ​​for all distant building groups within the target visual corridor; Extracting the inverted identification value sequence The median is used as the adaptive inverted baseline: ; in, Indicates the adaptive inverted reference value within the target's visual corridor; The following groups of buildings in the distance are defined as groups of buildings with inverted contrast between the distant background and the actual background: ,and: ; in, This indicates that the normalized contrast of the distant buildings is stronger than the normalized local color differences in the foreground, which is consistent with the inverted direction. This indicates that the inverted identification value of the building group in the distance is higher than the adaptive inverted benchmark in the target visual corridor, and it belongs to a building group with relatively prominent inverted features in the same target visual corridor. Grouping spatially adjacent groups of buildings with inverted background contrast into a controlled area for inverted background contrast; spatially adjacent means that two or more groups of buildings with inverted background contrast are adjacent in the horizontal direction of the distant view corridor image, or adjacent in the corresponding urban spatial unit. It should be noted that the inverted background contrast control zone refers to a zone consisting of one or more spatially adjacent groups of buildings with inverted background contrast. The buildings in this zone do not appear to have obvious local color abruptness in the close-up street view image, but they exhibit strong contrast prominence relative to the natural background in the distant view corridor image. The inverted background contrast control zone is used to indicate the urban space area that needs to be focused on for control in terms of building color, facade brightness, continuous interface effect, or view corridor background adaptation.

[0032] Specifically, when the zoning results are represented in the distant view corridor image, the horizontal projection interval of each distant background inverted building group in the distant view corridor image is used as the basis for adjacency judgment; if the horizontal projection intervals of two distant background inverted building groups are connected or overlapped, or if there is no distant building group corresponding to the conventional color coordination preservation area between them, then the two are determined to be spatially adjacent distant background inverted building groups; the distant building facade areas of the consecutively adjacent distant background inverted building groups are subjected to a union operation to obtain the zoning boundary of the distant background inverted control area in the distant view corridor image; When the zoning results are mapped to urban spatial units, the building outlines, building boundaries, or plot boundaries corresponding to each building object in the inverted background building group are combined to obtain the zoning boundary of the inverted background control area in the urban spatial base map. If the building outlines, building boundaries, or plot boundaries corresponding to two inverted background building groups are connected, or if they belong to the same continuous mountain interface, waterfront interface, or landscape corridor interface, then they are merged into the same inverted background control area. Thus, the zoning results can form a verifiable boundary expression in image coordinates and urban spatial coordinates. The remaining distant building groups that do not meet the criteria for inverted background contrast are grouped into the regular color coordination maintenance area. It should be noted that the regular color coordination maintenance area refers to distant building groups or continuous building areas that are not included in the distant background contrast inversion control area. The building groups in this area do not show obvious inversion characteristics of relatively harmonious foreground colors and prominent background contrast, and can be maintained or managed in accordance with the regular building color coordination requirements. The regular color coordination maintenance area does not mean that there are no other types of building color problems, but rather that it does not belong to the distant background contrast inversion type identified in this embodiment. Generate the architectural color and style zoning control results for the target visual corridor: ; in, This indicates the results of the zoning control of building color and style. This indicates a control area with inverted contrast between the distant view and the background. Indicates the area where regular color coordination is maintained; When outputting the results of building color and style zoning control, the zoning type, inverted identification value, corresponding building object number, and zoning boundary of each distant building group can be output. The zoning boundary can be represented by the building group boundary in the distant view corridor image, or it can be mapped to the building group boundary or control unit boundary in the urban spatial base map. The output results can be used for planning review, urban renewal style assessment, waterfront interface color control, mountainside building style control, or evaluation of the color coordination of buildings in important landscape corridors.

[0033] Example 2, as Figure 2 As shown, the machine learning-based urban building color and style zoning control system proposed in this invention is used to execute the machine learning-based urban building color and style zoning control method described in Embodiment 1, including: The image segmentation module is used to acquire the near-view street scene image and the far-view corridor image of the target corridor. It uses an image semantic segmentation model to extract the near-view building facade area, the far-view building facade area and the far-view natural background area, and determines the corresponding building objects between the near-view building facade area and the far-view building facade area. The foreground difference module is used to extract representative colors from the foreground building facade area and adjacent buildings, and calculate the foreground local color difference value of the building object. The distant contrast module is used to extract representative colors based on the distant building facade area and the adjacent distant natural background area, and to calculate the distant color contrast value of the building object; The distant view highlighting module is used to group continuously distributed distant building facade areas into distant building groups, and calculate the distant view contrast highlighting intensity by combining the distant view color contrast value and the horizontal projection continuity. The inverted recognition module is used to calculate the inverted recognition value of the distant building group based on the local color difference value of the foreground and the contrast intensity of the background of the corresponding building object. The partition control module is used to divide the distant background contrast inversion control area and the normal color coordination maintenance area based on the inverted recognition value.

[0034] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A machine learning-based method for zoning and controlling urban building color and style, characterized in that, Includes the following steps: S1. Obtain the near-view street scene image and the far-view corridor image of the target corridor. Use the image semantic segmentation model to extract the near-view building facade area, the far-view building facade area and the far-view natural background area, and determine the corresponding building objects between the near-view building facade area and the far-view building facade area. S2. Based on the representative colors extracted from the near-view building facade area and adjacent buildings, calculate the near-view local color difference value of the building object; S3. Extract representative colors based on the distant building facade area and the adjacent distant natural background area, and calculate the distant color contrast value of the building object; S4. Group the continuously distributed distant building facade areas into distant building groups, and calculate the distant contrast prominence intensity by combining the distant color contrast value and the horizontal projection continuity. S5. Based on the local color difference value of the foreground and the contrast intensity of the background of the corresponding building object, calculate the inverted recognition value of the background building group; S6. Based on the inverted recognition value, divide the distant background contrast inversion control area and the normal color coordination maintenance area.

2. The urban building color and style zoning control method based on machine learning according to claim 1, characterized in that, Extract the foreground building facade area, the background building facade area, and the background natural background area, including: A trained image semantic segmentation model is used to perform image semantic segmentation on the near-view street scene image and the far-view corridor image respectively, and extract the corresponding near-view building facade area, far-view building facade area and far-view natural background area. The distant natural background area includes at least one of mountains, water bodies, sky, and vegetation.

3. The urban building color and style zoning control method based on machine learning according to claim 1, characterized in that, The same building that appears simultaneously in both the close-up street view image and the distant view corridor image is identified as the corresponding building object.

4. The urban building color and style zoning control method based on machine learning according to claim 1, characterized in that, Calculating the foreground local color difference value of the building object includes: Convert the close-up street scene image to the Lab color space; The median of the lightness and chromaticity components of the pixels within the foreground building facade area in the Lab color space is extracted and used as the representative color of the foreground building. Identify at least one adjacent building that is horizontally adjacent to the building object in the close-up street view image, and use the median of the close-up building representative color of the adjacent building as the close-up local reference color; Calculate the Euclidean distance between the representative color of the foreground building and the local reference color of the foreground to obtain the local color difference value of the foreground.

5. The urban building color and style zoning control method based on machine learning according to claim 1, characterized in that, Calculating the background color contrast value of the building object includes: Convert the distant view corridor image to the Lab color space; The median of the pixels within the facade area of ​​the distant building is extracted in the Lab color space and used as the representative color of the distant building. Extract the area boundary of the distant building facade region; The boundary of the region is extended by adjacency according to the eight neighborhoods to obtain the boundary adjacency region surrounding the facade of the distant building. The intersection of the boundary adjacent region and the distant natural background region is performed to obtain the natural background pixels adjacent to the boundary of the distant building facade region. The median of the natural background pixels in the Lab color space is extracted as the representative color of the distant background; Calculate the Euclidean distance between the representative color of the distant building and the representative color of the distant background to obtain the color contrast value of the distant view.

6. The urban building color and style zoning control method based on machine learning according to claim 1, characterized in that, Calculating the contrast intensity of the distant view includes: In the distant view corridor image, with the same continuous distant natural background area as a reference, distant building facade areas with contiguous boundaries and continuous horizontal projection in the lateral direction are grouped into the distant building group. The horizontal projection continuity is calculated based on the coordinate range of the distant building group in the horizontal direction of the distant view corridor image and the distribution of building pixels. The overall color contrast of the building group is obtained by weighting the visible pixel area of ​​the building object in the distant view corridor image and calculating the distant color contrast value of the building object. The product of the horizontal projection continuity and the overall color contrast of the building group is used as the intensity of the background contrast.

7. The urban building color and style zoning control method based on machine learning according to claim 1, characterized in that, Calculating the inverted identification value of the vision building group includes: The foreground local color difference of the building object is obtained by weighting the visible pixel area of ​​the building object in the foreground street view image and calculating the foreground local color difference of the building object. Calculate the median and median absolute deviation of the distant contrast prominence intensity sequence for all distant building groups, and the median and median absolute deviation of the foreground local color difference sequence. Using the median and the absolute deviation of the median, adaptive normalization processing is performed on the contrast intensity of distant buildings and the local color difference of foreground buildings in each distant building group. The inverted recognition value is obtained by subtracting the normalized local color difference of the foreground from the normalized contrast intensity of the distant view.

8. The urban building color and style zoning control method based on machine learning according to claim 1, characterized in that, Based on the inverted recognition value, a distant background contrast inversion control area and a normal color coordination preservation area are divided, including: Construct an inverted identification value sequence for all the aforementioned vision building groups, and extract the median of the inverted identification value sequence as the adaptive inverted baseline value; The distant building group whose inverted identification value is greater than zero and greater than the adaptive inverted reference value is identified as the distant background contrast inverted building group. Grouping the spatially adjacent groups of inverted background buildings into a background contrast control zone, and grouping the remaining groups of background buildings into a regular color coordination maintenance zone.

9. The urban building color and style zoning control method based on machine learning according to claim 1, characterized in that, The same building that appears simultaneously in both the close-up street view image and the distant view corridor image is identified as the corresponding building object, including: Acquire at least one of the following: shooting location information, shooting direction information, and building area geometric information corresponding to the near-view street scene image and the far-view corridor image; Based on at least one of the shooting location information, shooting direction information, and building area geometric information, a candidate correspondence between the close-up building facade area in the close-up street view image and the distant building facade area in the distant view corridor image is determined. Extract the center position of the outline, the horizontal width ratio, and the vertical height ratio of the near-view building facade area and the far-view building facade area; Based on the candidate correspondence, the center position of the outline, the horizontal width ratio, and the vertical height ratio, the same building object corresponding to the near-view building facade area and the far-view building facade area is determined.

10. A machine learning-based urban building color and style zoning control system, used to execute the machine learning-based urban building color and style zoning control method according to any one of claims 1-9, characterized in that, Specifically, it includes: The image segmentation module is used to acquire the near-view street scene image and the far-view corridor image of the target corridor. It uses an image semantic segmentation model to extract the near-view building facade area, the far-view building facade area and the far-view natural background area, and determines the corresponding building objects between the near-view building facade area and the far-view building facade area. The foreground difference module is used to extract representative colors from the foreground building facade area and adjacent buildings, and calculate the foreground local color difference value of the building object. The distant contrast module is used to extract representative colors based on the distant building facade area and the adjacent distant natural background area, and to calculate the distant color contrast value of the building object; The distant view highlighting module is used to group continuously distributed distant building facade areas into distant building groups, and calculate the distant view contrast highlighting intensity by combining the distant view color contrast value and the horizontal projection continuity. The inverted recognition module is used to calculate the inverted recognition value of the distant building group based on the local color difference value of the foreground and the contrast intensity of the background of the corresponding building object. The partition control module is used to divide the distant background contrast inversion control area and the normal color coordination maintenance area based on the inverted recognition value.

Images