A method and system for street eye health assessment

Abstract

The application discloses a kind of street eye health assessment method and system, applied to eye health assessment technical field, method includes: to the multi-source data of target street is handled in time and space alignment, obtains multi-source feature dataset;Based on multi-source feature dataset, the target street is divided into first road section and second road section;Street view image is classified to each pixel semantics, and the street view visual index of first road section is obtained;The remote sensing image of second road section is input into the deep learning regression model trained, and the street view visual index of second road section is obtained;Obstruction contour data and air quality data are respectively correspondingly carried out visual analysis and spatial interpolation, and respectively correspondingly obtain visual distance index and air quality index;Street view visual index, visual distance index and air quality index are weighted and fused, and the eye health assessment result of target street is obtained.The application improves the evaluation accuracy of the influence of street environment on eye health.

Description

Technical Field

[0001] This invention relates to the field of eye health assessment technology, and in particular to a street-level eye health assessment method and system. Background Technology

[0002] Existing methods for assessing the impact of urban environments on eye health primarily involve: first, pre-constructing factors beneficial to eye health; then, assessing the quantity of these beneficial factors in different urban environments; and finally, determining the assessment result of the urban environment's impact on eye health based on the content of these beneficial factors. For example, given that natural greenery is known to be beneficial to eye health, the method assesses the proportion of natural greenery in urban park environments, which is higher than that in residential environments, thus outputting an assessment result that urban park environments are beneficial to eye health, while residential environments are not. However, existing methods assess only a single factor and cannot accurately assess the impact of street environments on eye health when faced with complex neighborhood environments. Summary of the Invention

[0003] This invention provides a method and system for assessing street eye health, in order to solve the technical problem that existing assessment methods for the impact of the street environment on eye health are not accurate enough.

[0004] To address the aforementioned technical problems, embodiments of the present invention provide a street-level eye health assessment method, comprising: Spatiotemporal alignment processing is performed on the multi-source image data, air quality data, and occlusion data of the target street to obtain the multi-source feature dataset of the target street. The target street is divided into road segments based on the multi-source feature dataset to obtain the first road segment and the second road segment of the target street; the multi-source feature dataset of the first road segment includes street view images, remote sensing images, occlusion outline data and air quality data; the multi-source feature dataset of the second road segment includes remote sensing images, occlusion outline data and air quality data. The street view image is subjected to pixel-by-pixel semantic classification to obtain the street view visual index of the first road segment; The remote sensing image of the second road segment is input into the trained deep learning regression model to obtain the street view visual index of the second road segment; the deep learning regression model is trained based on the street view visual index and remote sensing image of the first road segment. Visual analysis and spatial interpolation are performed on the outline data of the obstruction and the air quality data respectively to obtain the visual distance index and the air quality index respectively. The eye protection index is obtained by weighted fusion of the street view visual index, the visible distance index, and the air quality index. The eye protection index of the first road segment and the eye protection index of the second road segment are integrated to obtain the eye health assessment result of the target street.

[0005] As one preferred embodiment, the step of performing pixel-by-pixel semantic classification on the street view image to obtain the street view visual index of the first road segment includes: Extract the target street view image that meets the human eye's panoramic vision condition from the multi-source feature dataset of the first road segment; The target street view image is subjected to pixel-by-pixel semantic classification to obtain the semantic category of each effective pixel in the street view image; the semantic category includes vegetation and sky. Based on the effective pixel ratio of the vegetation, the green visibility index of the first road segment is determined; based on the effective pixel ratio of the sky, the sky visibility index of the first road segment is determined.

[0006] As one preferred embodiment, visual analysis is performed on the outline data of the obstruction to obtain a visual distance index, including: Determine the viewing angles of visible points on each segment of the target street; At the visible point, based on the outline data of the occluder within the visible angle, the visible directional distance from the visible point to the outline of the occluder is determined; Based on the visible direction distance, the visible distance index of each segment of the target street is obtained.

[0007] As one preferred embodiment, spatial interpolation is performed on the air quality data to obtain the air quality index, including: Determine the coordinates of the data collection points for the air quality data; When performing spatial interpolation on the air quality data of the first road segment, the spatial distance between the first road segment and the coordinates of the collection point is determined; Obtain the concentration of the target air pollutant in the air quality data, and the semi-variogram of the target air pollutant; Based on the spatial distance, the concentration, and the semi-variogram, the quality index of the target air pollutant in the first road segment is determined.

[0008] As one preferred embodiment, the step of performing pixel-by-pixel semantic classification on the street view image to obtain the street view visual index of the first road segment includes: Feature extraction is performed on the remote sensing image of the first road segment to obtain the remote sensing features of the first road segment; Based on the location data of the first road segment, the street view visual index and remote sensing features of the first road segment are paired point by point to obtain the training dataset and the validation dataset. A deep learning regression model is trained based on the training dataset and the validation dataset; the trained deep learning regression model is used to predict the street view visual index of the second road segment.

[0009] Another embodiment of the present invention provides a street eye health assessment system, comprising: The spatiotemporal alignment module is used to perform spatiotemporal alignment processing on the multi-source image data, air quality data and occlusion data of the target street to obtain the multi-source feature dataset of the target street. The road segmentation module is used to segment the target street based on the multi-source feature dataset to obtain a first road segment and a second road segment of the target street; the multi-source feature dataset of the first road segment includes street view images, remote sensing images, occlusion contour data and air quality data; the multi-source feature dataset of the second road segment includes remote sensing images, occlusion contour data and air quality data; A pixel-by-pixel semantic classification module is used to perform pixel-by-pixel semantic classification on the street view image to obtain the street view visual index of the first road segment; The model prediction module is used to input the remote sensing image of the second road segment into the trained deep learning regression model to obtain the street view visual index of the second road segment; the deep learning regression model is trained based on the street view visual index and remote sensing image of the first road segment. The data processing module is used to perform visual analysis and spatial interpolation on the outline data of the obstruction and the air quality data respectively, and obtain the visual distance index and air quality index respectively. The weighted fusion module is used to weight and fuse the street view visual index, the visible distance index, and the air quality index to obtain the eye protection index; The eye protection index integration module is used to integrate the eye protection index of the first road segment and the eye protection index of the second road segment to obtain the eye health assessment result of the target street.

[0010] As one preferred embodiment, the pixel-by-pixel semantic classification module includes: The target street view image extraction unit is used to extract the target street view image that meets the human eye's panoramic vision conditions from the multi-source feature dataset of the first road segment; A pixel-by-pixel semantic classification unit is used to perform pixel-by-pixel semantic classification on the target street view image to obtain the semantic category of each effective pixel of the street view image; the semantic category includes vegetation and sky; The street view visual index determination unit is used to determine the green view rate index of the first road segment based on the effective pixel ratio of the vegetation; and to determine the sky view index of the first road segment based on the effective pixel ratio of the sky.

[0011] As one preferred embodiment, the data processing module includes: A viewing angle determination unit is used to determine the viewing angle of a visible point on each segment of the target street. A visible direction distance determination unit is used to determine the visible direction distance from the visible point to the outline of the occluder based on the outline data of the occluder within the visible angle at the visible point. The visible distance index determination unit is used to obtain the visible distance index of each segment of the target street based on the visible direction distance.

[0012] As one preferred embodiment, the data processing module further includes: The unit for determining the coordinates of the collection points for the air quality data is used to determine the coordinates of the collection points. The spatial distance determination unit is used to determine the spatial distance between the first road segment and the coordinates of the collection point when performing spatial interpolation on the air quality data of the first road segment. A semi-variogram determination unit is used to obtain the concentration of the target air pollutant in the air quality data, and the semi-variogram of the target air pollutant; An air quality index determination unit is used to determine the air quality index of the target air pollutant in the first road segment based on the spatial distance, the concentration, and the semi-variogram.

[0013] As one preferred embodiment, the street eye health assessment system further includes: The remote sensing feature extraction module is used to extract features from the remote sensing image of the first road segment to obtain the remote sensing features of the first road segment. The dataset construction module is used to perform point-by-point pairing of the street view visual index and remote sensing features of the first road segment based on the positioning data of the first road segment, so as to obtain the training dataset and the validation dataset. The model training module is used to train a deep learning regression model based on the training dataset and the validation dataset; the trained deep learning regression model is used to predict the street view visual index of the second road segment.

[0014] This application achieves comprehensive coverage and refined quantitative assessment of the street-level eye health environment by combining street scene semantic segmentation, remote sensing regression inference, visual distance analysis, and air quality spatial interpolation through multi-source data spatiotemporal alignment and road segment division. A deep learning regression model is trained using ground truth street scene data from the first road segment, effectively compensating for the sparseness of street scene sampling points and enabling reliable estimations of visual environment factors even for the second road segment without street scene coverage. Based on this, a unified eye protection index is generated by fusing multi-dimensional visual and air factors and weighting them according to evidence-based data, thus improving the accuracy of assessing the impact of the street environment on eye health. Attached Figure Description

[0015] Figure 1 This is one of the flowcharts of the street eye health assessment method provided by the present invention; Figure 2 This is the second flowchart of the street eye health assessment method provided by the present invention; Figure 3 This is a schematic diagram of the structure of the street eye health assessment system provided by the present invention.

[0016] Figure label: Among them, 301 is the spatiotemporal alignment module; 302 is the road segment division module; 303 is the pixel-by-pixel semantic classification module; 304 is the model prediction module; 305 is the data processing module; 306 is the weighted fusion module; and 307 is the eye protection index integration module. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0019] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0020] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0021] See Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the street eye health assessment method provided by the present invention, as shown below. Figure 1 As shown, this embodiment includes steps 100 to 700, and the specific steps are as follows: Step 100: Perform spatiotemporal alignment processing on the multi-source image data, air quality data, and occlusion data of the target street to obtain the multi-source feature dataset of the target street; In this embodiment, the multi-source image data refers to image data obtained from different data sources that can reflect the characteristics of the street spatial environment, including satellite remote sensing imagery and street view images. The air quality data in this embodiment refers to air pollutant concentration data collected in real-time or periodically by ground-based air quality monitoring stations, including concentration values ​​of common pollutants such as PM2.5, PM10, carbon monoxide (CO), ozone (O3), nitrogen dioxide (NO2), and sulfur dioxide (SO2). The obstruction contour data in this embodiment refers to data describing the three-dimensional spatial morphology of entities such as buildings and structures surrounding the street, including three-dimensional building contour data.

[0022] The purpose of the spatiotemporal alignment processing in this embodiment is to unify multi-source data from different sources with varying spatiotemporal references under the same geographic coordinate system and time reference frame. This eliminates deviations caused by inconsistencies in coordinate systems and acquisition times between different data sources, ensuring that subsequent multi-factor (also known as index) aggregation and calculations based on road segments can be processed at the same spatial location and temporal dimension. Specifically, this can be achieved by: performing geographic coordinate registration on satellite remote sensing imagery, street view panoramic images, air quality monitoring data, and 3D building outline data, and uniformly converting them to the WGS-84 or CGCS-2000 coordinate system; for the time dimension, aligning the timestamps of each data source according to the acquisition time, and grouping multiple data points collected for the same road segment within the same time period into the same spatiotemporal data record.

[0023] After the above spatiotemporal alignment process, a multi-source feature dataset of the target street can be obtained. This dataset is indexed by unified geographic coordinates and time labels and contains image features, air quality features and occlusion morphology features corresponding to the spatial location of each road segment of the target street, providing a data foundation for subsequent road segment division and calculation of eye health influencing factors.

[0024] In one specific implementation of this embodiment, the spatiotemporal alignment processing can be performed as follows: First, the geographical range of the target street is obtained, and the road network data within this range is determined. Then, using nodes and road segments in the road network as spatial indices, satellite remote sensing images are extracted to the corresponding road segment buffers according to pixel positions. Street view images are matched to the nearest road segment according to the latitude and longitude coordinates of the acquisition points. Air quality monitoring data is allocated to each road segment according to the monitoring station locations using spatial interpolation methods. Three-dimensional building outline data is mapped to the corresponding road segment areas according to the spatial range of the building footprint. Through the above methods, each road segment obtains a set of multi-source feature data that is spatially aligned and temporally matched.

[0025] By using spatiotemporal alignment processing, multiple data sources with scattered origins and different formats are unified and integrated into a multi-source feature dataset that is spatially accurate and temporally comparable. This lays the data foundation for subsequent eye health assessments based on road segments as the basic unit, ensuring the spatial accuracy and spatiotemporal consistency of the assessment results.

[0026] Step 200: Divide the target street into road segments based on the multi-source feature dataset to obtain the first road segment and the second road segment of the target street; the multi-source feature dataset of the first road segment includes street view images, remote sensing images, occlusion contour data and air quality data; the multi-source feature dataset of the second road segment includes remote sensing images, occlusion contour data and air quality data. In this embodiment, a road segment refers to a basic spatial unit obtained by dividing the road network according to certain rules. It is the smallest spatial granularity for subsequent aggregation and calculation of various environmental factors. The division of road segments can be based on the topology, node distribution, or equidistant intervals of the road network. In this embodiment, the first road segment refers to a road segment in the multi-source feature dataset that simultaneously contains street view images, remote sensing images, occlusion contour data, and air quality data. In this embodiment, the second road segment refers to a road segment in the multi-source feature dataset that contains remote sensing images, occlusion contour data, and air quality data, but does not contain street view images. That is, the first road segment is a road segment covered by street view images, and the second road segment is a road segment not covered by street view images.

[0027] In this embodiment, the specific method for road segment division can be as follows: acquire road network data of the target street, and based on the road network, use road intersections, road start points, and road end points as nodes, and use the road segments between adjacent nodes as initial road segments. For road segments that are too long, they can be further divided into finer-grained sub-segments according to preset equal intervals to ensure that the spatial resolution of the evaluation meets the requirements for refinement.

[0028] After the road segmentation is completed, for each road segment, it is checked whether its multi-source feature dataset contains street view image data. If a road segment's multi-source feature dataset contains street view images, then the road segment is marked as the first road segment; if a road segment's multi-source feature dataset does not contain street view images, then the road segment is marked as the second road segment. Thus, all road segments of the target street are divided into the first road segment and the second road segment. The visual environment factors for the first road segment can be directly extracted from street view images, while the visual environment factors for the second road segment need to be inferred using remote sensing data.

[0029] In one specific implementation of this embodiment, road network data can be projected and stored in advance according to unified geographic coordinates. The results of road segment division can be stored in the geographic information system in the form of vector line segments. Each road segment is assigned a unique road segment identifier and associated with the spatial geometric information of the road segment (such as starting coordinates, ending coordinates, road segment length, direction, etc.) and the label of the road segment category (first road segment or second road segment).

[0030] By finely dividing the target streets into road segments and classifying them into first and second road segments based on the presence or absence of street view images, a structured data organization foundation is provided for subsequent use of differentiated visual factor extraction strategies for different road segment types. This allows road segments with street view coverage to obtain visual ground truth through high-precision semantic segmentation, while road segments without street view coverage can be supplemented and estimated through remote sensing inference, thereby achieving full street coverage while ensuring evaluation accuracy.

[0031] Step 300: Perform pixel-by-pixel semantic classification on the street view image to obtain the street view visual index of the first road segment; In this embodiment, the street view visual index refers to a quantitative indicator obtained by performing pixel-by-pixel semantic classification on street view images and then statistically analyzing the proportion of pixels in each category. It is used to characterize the features of the street visual environment in a certain dimension. The street view visual index may include green view rate index, sky visibility index, street enclosure index, and traffic road proportion index, etc.

[0032] The pixel-by-pixel semantic classification in this embodiment is an image processing operation that assigns a semantic category label to each pixel in a street view image. This operation identifies the categories and spatial distribution of different land features in the street view image, thus providing a basis for subsequent statistical analysis of the pixel proportions of each category. Specifically, a deep learning-based semantic segmentation model can be used to process the street view image. The deep learning semantic segmentation model performs pixel-by-pixel classification on the street view image, classifying each pixel into one of the following semantic categories: vegetation, sky, buildings, walls, fences, roads, pedestrians, and vehicles.

[0033] In one specific implementation of this embodiment, the street view image can be preprocessed before being input into the semantic segmentation model, including: cropping or projecting the panoramic street view image according to the normal viewing angle range of the human eye to extract the image region that conforms to the actual perceptual field of vision of the human eye; normalizing the image size to make it meet the input size requirements of the semantic segmentation model; and performing color space conversion or normalization on the image to eliminate the influence of lighting differences under different acquisition conditions on the segmentation accuracy.

[0034] After the semantic segmentation model performs pixel-by-pixel classification of the street view image, it counts the number of pixels in each semantic category within the effective pixel area of ​​the image. Let N_vegetation represent the number of pixels classified as vegetation; N_sky represent the number of pixels classified as sky; N_building represent the number of pixels classified as building; N_wall represent the number of pixels classified as wall; N_fence represent the number of pixels classified as fence; N_road represent the number of pixels classified as road; and N_total represent the total number of effective pixels in the image. Then, the green view rate index can be expressed as N_vegetation ÷ N_total, the sky view rate index can be expressed as N_sky ÷ N_total, the street enclosure index can be expressed as (N_building + N_wall + N_fence) ÷ N_total, and the traffic road ratio index can be expressed as N_road ÷ N_total. These indices constitute the street view visual indices of the first road segment.

[0035] In another implementation of this embodiment, the street view image can be defined as an image collected within the pedestrian area of ​​the street, along the normal viewing angle range of the human eye, to conform to the actual perception scenario of eye health. Specifically, an image area within the range of 0° to 360° in the horizontal direction and -30° to +30° in the vertical direction can be cropped from the street view image as the effective analysis area. This angle range corresponds to the viewing angle range of a pedestrian's eyes when walking normally.

[0036] By performing pixel-by-pixel semantic classification on street view images, the proportion of various elements closely related to eye health in the street visual environment can be accurately quantified at the pixel level, such as green coverage (green view rate), openness (sky visibility), spatial enclosure (street enclosure), and the proportion of motor vehicle traffic space (traffic road proportion), providing refined quantitative indicators of the visual environment for eye health assessment.

[0037] Step 400: Input the remote sensing image of the second road segment into the trained deep learning regression model to obtain the street view visual index of the second road segment; the deep learning regression model is trained based on the street view visual index and remote sensing image of the first road segment. The deep learning regression model in this embodiment is a regression prediction model built on a deep neural network architecture, used to establish a mapping relationship between remote sensing image features and street view visual indices. The model's function is to estimate the street view visual index of a second road segment lacking street view images using only remote sensing images, thereby achieving continuous coverage of the street view visual index across the entire street area.

[0038] In this embodiment, remote sensing images refer to long-range perception images acquired from satellites or airborne platforms, which may include synthetic aperture radar images, multispectral images, and high-resolution optical images. This embodiment uses a deep learning regression model, taking data from the first road segment that simultaneously possesses both ground truth street view visual index values ​​and remote sensing image features as training samples, to learn the complex mapping relationship between remote sensing image features and the street view visual index. The trained model is then applied to the second road segment, which only contains remote sensing images, to infer the street view visual index of the second road segment.

[0039] In one specific implementation of this embodiment, the training process of the deep learning regression model is as follows: First, feature extraction is performed on the remote sensing image of the first road segment to obtain the remote sensing features of the first road segment. The remote sensing features may include the backscattering coefficient of satellite synthetic aperture radar, the reflectance of each band of satellite multispectral imagery, the RGB band values ​​of satellite seasonal synthetic imagery, and depth maps, perception intensity, and normalized vegetation index derived from the above images. Second, based on the geographical location information of the first road segment, the street view visual index of the first road segment is paired point-by-point with the remote sensing features according to the same spatial location to form a training dataset and a validation dataset. Then, using the remote sensing features as input and the corresponding street view visual index as the supervision label, the deep learning regression model is trained. The model training can use mean squared error (MSE) as the loss function and employ optimization algorithms such as stochastic gradient descent or Adam for parameter updates. After training, the remote sensing features of the second road segment are input into the model, and the model output is the estimated value of the street view visual index of the second road segment.

[0040] By using a deep learning regression model, visual environmental factors in areas not covered by street view are estimated by inferring from remote sensing image features. This effectively overcomes the problem that street view images only cover a limited number of sampling points and are difficult to achieve full city coverage, enabling the evaluation results to cover all street areas.

[0041] Step 500: Perform visual analysis and spatial interpolation on the outline data of the obstruction and the air quality data respectively to obtain the visual distance index and air quality index respectively. In this embodiment, the visual distance index is a quantitative indicator that characterizes the distance at which a person's eye can see in the distance from a certain location on a street. This index is calculated based on the outline data of obstructions and is used to reflect the openness of the visual environment at that location. A greater viewing distance is more beneficial to eye health; therefore, the visual distance index is a positive factor in eye health assessment.

[0042] The visual analysis in this embodiment is a spatial analysis operation used to calculate the visible distance in the directions of the viewpoint based on the contour data of three-dimensional occlusion objects. The purpose of this operation is to quantify the distances that a human eye can see when looking around from a certain location. Specifically, at the road sampling point, multiple rays can be emitted at equal angles along multiple directions of the human eye's gaze. The distance between each ray and the nearest intersection point with the boundaries of surrounding buildings is calculated, and the average of these distances is taken as the visible distance at that viewpoint. An upper limit can be set for the visible distance; distances exceeding this upper limit are counted according to the upper limit value.

[0043] In one specific implementation of this embodiment, the calculation steps for the visibility distance index are as follows: First, acquire the outline data of obstructions on the target street; the outline data can be three-dimensional building outline data. Second, set visibility points on the road network according to a preset sampling interval; the visibility points can be located at the sidewalk. Then, at each visibility point, with that point as the origin, emit a ray at regular intervals in the horizontal direction from 0° to 360°, and in the vertical direction, consider the normal eye-level viewing angle (e.g., horizontal elevation angle 0° to 10°). For each ray, calculate its intersection with the outlines of surrounding buildings, and record the distance from the intersection to the origin. If the ray does not intersect with any building within a set upper limit distance, the distance in that direction is recorded as the upper limit value. Finally, average the distances in all directions to obtain the visibility distance of the visibility point. Buffering and aggregating the visibility distances of the visibility points along the road segment centerline yields the street-level visibility distance index.

[0044] In this embodiment, the Air Quality Index (AQI) is a quantitative indicator characterizing the degree of air pollution at a specific location on a street. This index is calculated through spatial interpolation based on monitoring data from ground-based air quality monitoring stations, reflecting the combined concentration levels of various air pollutants at that monitoring station location. Higher concentrations of air pollutants have a greater impact on eye health; therefore, all pollutant indicators in the AQI are negative factors for eye health assessment.

[0045] Spatial interpolation in this embodiment is a spatial data processing operation used to estimate pollutant concentrations at unmonitored locations based on pollutant concentration data from discretely distributed monitoring stations. This operation expands sparse station monitoring data into spatially continuous data covering the entire area. Specifically, ordinary kriging can be used for spatial interpolation. Ordinary kriging is a geostatistical spatial interpolation method that assigns different weights to each sample point based on their spatial relationships and correlations, achieving optimal unbiased estimation of unknown point values ​​through a moving weighted average.

[0046] In one specific implementation of this embodiment, the calculation steps for the air quality index are as follows: First, obtain the location coordinates of various surface air quality monitoring stations within the target street area and the concentration data of multiple pollutants collected by each station. Second, for each pollutant, establish a semi-variogram model, for example, using a spherical variogram model, to describe the spatial correlation and variability of the pollutant concentration. Then, for each road segment on the target street, based on the spatial coordinates of the center point of the segment, use ordinary kriging, based on the spatial distance between the point and each monitoring station and the semi-variogram model, to calculate the weight of each monitoring station's data at that point, and sum the weighted values ​​to obtain the estimated concentration of the pollutant at that point. Finally, convert the estimated concentration values ​​of each pollutant into scores according to a preset standardization method to obtain the air quality index for each pollutant. By weightedly combining the air quality indices of multiple pollutants, a comprehensive air quality index can be obtained.

[0047] Visual analytics transforms 3D building outline data into a visual distance index, quantifying the openness of the street's visual environment—a spatial characteristic closely related to eye health. Spatial interpolation transforms discrete air quality data from monitoring stations into a continuous spatial distribution, enabling each street to receive a corresponding air quality assessment. The combination of these two methods achieves a comprehensive quantification of the street environment from both visual and air quality dimensions.

[0048] Step 600: The street view visual index, visible distance index and air quality index are weighted and fused to obtain the eye protection index; The eye protection index in this embodiment refers to a quantitative scoring index that comprehensively reflects the degree of eye health protection provided by the street environment. This index integrates multiple dimensions of assessment indicators from two major categories: visual environment factors and air pollution factors. The value range can be set from 0 to 100 points. The higher the score, the more conducive the street environment is to eye health, and the lower the score, the greater the potential risk to eye health from the street environment.

[0049] The weighted fusion in this embodiment is a data processing operation used to comprehensively calculate the factor scores from multiple dimensions according to their respective weights, resulting in a single comprehensive score index. The purpose of this operation is to integrate multi-dimensional environmental assessment results into a single numerical value, facilitating subsequent applications.

[0050] In one specific implementation of this embodiment, the calculation steps for the eye protection index are as follows: First, the original values ​​of each factor are standardized and uniformly mapped to a score range of 0 to 100. To suppress the influence of extreme values ​​on the standardization results, the 1st to 99th percentiles of each factor can be taken as the clipping boundaries for the minimum and maximum values ​​across the entire city. Values ​​exceeding this range are treated as boundary values.

[0051] Next, assign weights to each factor. For example, the green visibility factor can be assigned a weight of 17, the PM2.5 factor 14, the sky visibility factor 13, the O3 factor 11, the visibility distance factor 10, the NO2 factor 8.5, the PM10 factor 7, the street enclosure factor 6, the SO2 factor 6.5, the traffic road ratio factor 4, and the CO factor 3. These weight values ​​are for illustrative purposes only; in actual applications, the weights are configurable parameters and can be adjusted.

[0052] Then, the eye protection index is calculated according to the weighted summation formula. , ,in, Let i be the weight of the i-th factor. This is the standardized score for this factor.

[0053] By weighting and integrating multi-dimensional and multi-factor environmental assessment indicators, a comprehensive eye protection index is obtained. This index not only fully reflects various factors related to eye health in the street environment, but also reflects the differences in the degree of impact of different factors on eye health through evidence-based weights, making the assessment results more accurate.

[0054] Step 700: Integrate the eye protection index of the first road segment and the eye protection index of the second road segment to obtain the eye health assessment result of the target street.

[0055] In this embodiment, the eye health assessment result refers to the complete assessment dataset obtained after calculating the eye protection index for all road segments of the target street. This dataset contains the eye protection index value and corresponding spatial location information for each road segment within the target street.

[0056] In this embodiment, integration refers to the data processing operation of merging the eye protection indices of the first and second road segments, calculated separately, according to the spatial order and topological relationship of the road segments, to form a complete street network eye protection index dataset. The purpose of the integration operation is to aggregate the scattered road segment evaluation results into a unified and complete street-level evaluation result.

[0057] In one specific implementation of this embodiment, the integration of eye health assessment results can be achieved through the following steps: First, obtain a list of road segment identifiers and their spatial geometric information for all road segments of the target street. Second, for each road segment in the first road segment set, associate its corresponding eye protection index value with the road segment identifier; for each road segment in the second road segment set, associate its corresponding eye protection index value with the road segment identifier. Then, merge the eye protection index values ​​of all road segments according to their road segment identifiers to form a complete eye protection index dataset. Finally, the dataset can be classified according to the level of the eye protection index, for example, into three levels: high risk (0-33 points), medium risk (34-66 points), and low risk (67-100 points), and the classification results can be associated with the spatial geometric information of the road segments to generate vector data that can be used for map rendering.

[0058] By integrating the eye protection indices of the first and second road sections, a complete eye health assessment result covering all road sections of the target street was formed, making the assessment results have full coverage and spatial integrity.

[0059] This application achieves comprehensive coverage and refined quantitative assessment of the street-level eye health environment by combining street scene semantic segmentation, remote sensing regression inference, visual distance analysis, and air quality spatial interpolation through multi-source data spatiotemporal alignment and road segment division. A deep learning regression model is trained using ground truth street scene data from the first road segment, effectively compensating for the sparseness of street scene sampling points and enabling reliable estimations of visual environment factors even for the second road segment without street scene coverage. Based on this, a unified eye protection index is generated by fusing multi-dimensional visual and air factors and weighting them according to evidence-based data, thus improving the accuracy of assessing the impact of the street environment on eye health.

[0060] See Figure 2 , Figure 2 This is a flowchart illustrating another embodiment of the street eye health assessment method provided by the present invention, as shown below. Figure 2 As shown, this embodiment includes steps 310 to 330, and the specific steps are as follows: Step 310: Extract the target street view image that meets the human eye's panoramic vision conditions from the multi-source feature dataset of the first road segment; Step 320: Perform pixel-by-pixel semantic classification on the target street view image to obtain the semantic category of each effective pixel of the street view image; the semantic category includes vegetation and sky; Step 330: Based on the effective pixel ratio of the vegetation, determine the green visibility index of the first road segment; based on the effective pixel ratio of the sky, determine the sky visibility index of the first road segment.

[0061] First, target street view images conforming to the human eye's panoramic vision conditions are extracted from the multi-source feature dataset of the first road segment. In this embodiment, the human eye's panoramic vision conditions refer to the visual field and orientation of the image acquisition matching the natural panoramic vision conditions of a pedestrian walking normally. Street view images are typically panoramic images, but not all visual field ranges are relevant to the actual perception of eye health. Therefore, it is necessary to extract image regions from the panoramic images that conform to the normal panoramic vision angle range of the human eye as target street view images. Specifically, image regions within the horizontal range of 0° to 360° and the vertical range of -30° to +30° can be extracted. This angle range corresponds to the natural field of vision of a pedestrian walking normally, including horizontal panoramic vision and slightly upward and downward views.

[0062] Secondly, pixel-by-pixel semantic classification is performed on the target street view image to obtain the semantic category of each effective pixel in the street view image. The semantic categories include at least vegetation and sky. In this embodiment, the semantic category refers to the category label assigned after classifying each pixel in the image using a semantic segmentation model, and can include multiple categories such as vegetation, sky, buildings, walls, fences, roads, pedestrians, and vehicles. Pixel-by-pixel semantic classification can be implemented using a deep learning semantic segmentation model.

[0063] Then, based on the effective pixel proportion of vegetation, the green view rate index of the first road segment is determined. Specifically, the number of pixels classified as vegetation (N_vegetation) in the target street view image is counted, and the total number of effective pixels in the image (N_total) is counted. The green view rate index = N_vegetation ÷ N_total. The green view rate index reflects the proportion of street greenery in the pedestrian's field of vision and is an important indicator for measuring the visual exposure level of street greenery.

[0064] Based on the effective pixel ratio of the sky, the sky visibility index of the first road segment is determined. Specifically, the number of pixels classified as sky in the target street view image, N_sky, is counted, and the sky visibility index = N_sky ÷ N_total. The sky visibility index reflects the openness of the street space; the higher the sky ratio, the wider the view and the more abundant the lighting.

[0065] In one specific implementation of this embodiment, the semantic segmentation model can further classify pixels into more categories such as buildings, walls, fences, and roads, thereby enabling the calculation of the street enclosure index and the traffic road proportion index. The street enclosure index = (N_buildings + N_walls + N_fences) ÷ N_total, used to characterize the degree of enclosure of buildings on both sides of the street; the traffic road proportion index = N_roads ÷ N_total, used to characterize the proportion of motor vehicle traffic space in the field of view. Both street enclosure and traffic road proportion are negative harmful factors and can be assigned inversely.

[0066] This embodiment extracts target street scene images that meet the conditions of human eye panoramic vision and performs pixel-by-pixel semantic classification. It can accurately quantify the two visual factors most closely related to eye health in the street visual environment: vegetation coverage (green view rate) and sky openness (sky visible area). This makes the evaluation indicators truly start from the perspective of human eye perception and fit the actual assessment needs of eye health. At the same time, the semantic segmentation method based on deep learning ensures the accuracy of classification.

[0067] In this embodiment, the green view rate index and the sky visibility index, along with the optional street enclosure index and traffic road ratio index, together constitute the street view visual index of the first road segment, providing basic data for the visual dimension of the subsequent eye protection index calculation.

[0068] In another embodiment of the street eye health assessment method provided by the present invention, the above steps specifically include: Step 510: Determine the viewing angle of the visible points on each section of the target street; Step 520: At the visible point, based on the outline data of the occluder within the visible angle, determine the visible directional distance from the visible point to the outline of the occluder. Step 530: Based on the visible direction distance, obtain the visible distance index of each segment of the target street.

[0069] First, determine the viewing angles of visible points on each segment of the target street. In this embodiment, a visible point refers to a spatial location on the road network selected according to a preset sampling interval for calculating the viewing distance. Visible points can be located on pedestrian walkways, and their specific locations can be determined based on the actual needs of the evaluation. In this embodiment, the viewing angle refers to the set of directional angles of rays emitted from a viewpoint during visual analysis. The viewing angle can include multiple azimuth angles in the horizontal direction and multiple elevation angles in the vertical direction.

[0070] In one specific implementation of this embodiment, an azimuth angle can be set every 15° from 0° to 360° in the horizontal direction, for a total of 24 directions; in the vertical direction, the eye-level viewing angle can be mainly considered, i.e., the elevation angle range of 0° to 10°, or only two-dimensional visual analysis in the horizontal direction can be used. The setting of the viewing angle should simulate the natural panoramic viewing angle range of the human eye to ensure that the calculated viewing distance can reflect the actual distance that the human eye can see.

[0071] Secondly, at the viewpoint, based on the outline data of the obstruction within the view angle, the directional distance from the viewpoint to the outline of the obstruction is determined. In this embodiment, the outline data of the obstruction refers to data describing the three-dimensional spatial shape and location of obstructions such as buildings and structures around the street, which may include the planar footprint and height information of the buildings. The directional distance in this embodiment refers to the straight-line distance from the viewpoint along a specific direction to the outline of the first obstruction.

[0072] In one specific implementation of this embodiment, the visible direction distance can be calculated using a ray intersection method: From a certain viewpoint, a ray is emitted along a certain view angle direction, extending outward from the viewpoint. The intersection of this ray with the three-dimensional contours of all surrounding buildings is calculated, and the first intersection point is found. The distance from this intersection point to the viewpoint is recorded. If the ray does not intersect with any building within a preset maximum search distance, the visible direction distance in that direction is recorded as the maximum search distance value (e.g., 500 meters). This calculation is performed for each view angle direction of the viewpoint to obtain the set of visible direction distances for that viewpoint in each direction.

[0073] Then, based on the visible directional distances, the visible distance indices for the first and second road segments are obtained. Specifically, for a visible point, the average of its visible directional distances in all directions is taken to obtain the visible distance of that point. For a road segment, the visible distances of all visible points on that segment are spatially aggregated along the centerline of the segment (e.g., by taking the average or median) to obtain the visible distance index of that road segment.

[0074] In one specific implementation of this embodiment, the calculation of the visual distance index can also take into account the differences in importance of different directions. For example, a higher weight can be given to the pedestrian's forward direction (front and back direction), and a lower weight can be given to the left and right sides, so as to more accurately reflect the pedestrian's visual experience during actual walking.

[0075] This embodiment uses the ray intersection method based on three-dimensional occlusion contour data for visual analysis, which can accurately quantify the distance that the human eye can see in the distance at various locations on the street. This indicator directly reflects the openness of the street visual environment and the conditions for looking into the distance, which is in line with the recommendations for eye health care and provides a visual environment indicator with clear physiological significance for eye health assessment.

[0076] In another embodiment of the street eye health assessment method provided by the present invention, the above steps specifically include: Step 540: Determine the coordinates of the air quality data collection points; Step 550: When performing spatial interpolation on the air quality data of the first road segment, determine the spatial distance between the first road segment and the coordinates of the collection point; Step 560: Obtain the concentration of the target air pollutant in the air quality data, and the semi-variogram of the target air pollutant; Step 570: Based on the spatial distance, the concentration, and the semi-variogram, determine the quality index of the target air pollutant in the first road segment.

[0077] First, the coordinates of the air quality data collection points are determined. In this embodiment, the collection point coordinates refer to the geographical coordinates of the ground-based air quality monitoring stations. Second, when performing spatial interpolation on the air quality data of the first road segment, the spatial distance between the first road segment and the collection point coordinates is determined. In this embodiment, the spatial distance refers to the geographic spatial distance between the center point of the road segment or the road segment area and each air quality monitoring station, usually measured by Euclidean distance or geodetic distance. Spatial distance is an important basis for determining the weight of each monitoring station in ordinary Kriging interpolation—the closer the monitoring station is to the unknown point, the greater its contribution to the estimate of the unknown point.

[0078] Then, the concentration of the target air pollutant and its semivariogram are obtained from the air quality data. In this embodiment, the target air pollutant can be any one or more of PM2.5, PM10, CO, O3, NO2, and SO2. The semivariogram (also known as the semivariogram function) in this embodiment is a function used to describe the spatial correlation of regionalized variables; it reflects the way and extent to which variable values ​​change with increasing spatial distance. The semivariogram is typically characterized by three parameters: sill value, nugget value, and range. Different pollutants have different spatial distribution characteristics; therefore, corresponding semivariogram models need to be established separately.

[0079] Then, based on spatial distance, concentration, and semivariogram, the air quality index of the target air pollutant for the first road segment is determined. Specifically, for a given location on the first road segment, ordinary kriging is used for interpolation estimation: based on the spatial distance between this point and each monitoring station, and combined with the semivariogram model of the target pollutant, the weighting coefficients of the data from each monitoring station at that point are calculated to minimize the variance of the estimated value and ensure unbiasedness. The pollutant concentration values ​​from each monitoring station are then weighted and summed with their corresponding weighting coefficients to obtain the estimated concentration of the pollutant at that point.

[0080] In one specific implementation of this embodiment, for each pollutant, the above interpolation process is repeated to obtain the concentration estimate of each pollutant at that road segment. Then, the concentration estimate of each pollutant is converted according to a preset standardization method, for example, by using the min-max normalization method to map it to a score range of 0 to 100. For pollutants, the higher the concentration, the lower the score, i.e., reverse scoring, to obtain the quality index of each pollutant.

[0081] This embodiment uses ordinary kriging spatial interpolation to transform the air quality data from discrete monitoring stations into continuous spatial distribution data covering all road segments, enabling each street to obtain a corresponding air quality assessment. Ordinary kriging considers the spatial correlation and structure of pollutant concentrations. By establishing semi-variogram models for different pollutants, the spatial distribution characteristics of different pollutants can be more accurately characterized.

[0082] In another embodiment of the street eye health assessment method provided by the present invention, the above steps specifically include: Step 10: Extract features from the remote sensing image of the first road segment to obtain the remote sensing features of the first road segment; Step 20: Based on the location data of the first road segment, perform point-by-point pairing of the street view visual index and remote sensing features of the first road segment to obtain the training dataset and the validation dataset. Step 30: Train a deep learning regression model based on the training dataset and the validation dataset; the trained deep learning regression model is used to predict the street view visual index of the second road segment.

[0083] After obtaining the street view visual index of the first road segment, the remote sensing image of the first road segment is first subjected to feature extraction to obtain the remote sensing features of the first road segment. In this embodiment, the remote sensing features refer to quantitative indicators extracted from the remote sensing image that reflect environmental information such as land cover type, vegetation status, and building density. In this embodiment, the location data refers to the geospatial location information of the first road segment, which can be the coordinates of the center point of the road segment. The role of the location data is to associate the street view visual index and the remote sensing features at the same spatial location, forming a one-to-one corresponding training sample pair. The specific method of point-by-point pairing is as follows: For each first road segment, the location data of that road segment is obtained. Using this location data as an index, the street view visual index value of that road segment is extracted from the street view visual index dataset, and the remote sensing feature vector corresponding to that location is extracted from the remote sensing feature dataset. The two are then paired as a training sample. All paired samples of the first road segments are randomly divided into training datasets and validation datasets according to a preset ratio (e.g., 7:3 or 8:2).

[0084] Then, a deep learning regression model is trained using the training and validation datasets. In this embodiment, the deep learning regression model employs a network architecture combining convolutional neural networks (CNNs) and fully connected layers. The convolutional layers extract high-level spatial features from remote sensing data, while the fully connected layers map the extracted features to continuous numerical outputs (i.e., estimates of the street view visual index). During model training, remote sensing feature vectors are used as input, the corresponding ground truth street view visual index values ​​are used as supervision labels, mean squared error (MSE) is used as the loss function, and backpropagation and an optimizer (such as Adam) are employed to iteratively update the network parameters. During training, the model is validated using a validation dataset, and the loss on the validation set is monitored. Training stops when the validation set loss no longer decreases to prevent overfitting.

[0085] The trained deep learning regression model is used to predict the street view visual index of the second road segment. Specifically, the remote sensing feature vector of the second road segment is input into the trained model, and the model output is the estimated value of the street view visual index of that road segment.

[0086] This embodiment constructs a training dataset by pairing the ground truth values ​​of the street view visual index of the first road segment with remote sensing features point by point, and trains a deep learning regression model, enabling the model to learn the complex mapping relationship between remote sensing image features and the street view visual index. The trained model can be applied to the second road segment to expand the street view visual index from sparse sampling to full coverage. As new street view data is continuously added, the model can be continuously iterated and updated to continuously improve the inference accuracy.

[0087] The street eye health assessment system provided by the present invention is described below. The street eye health assessment system described below can be referred to in correspondence with the street eye health assessment method described above.

[0088] Please refer to Figure 3 The present invention also provides a street eye health assessment system, comprising: The spatiotemporal alignment module 301 is used to perform spatiotemporal alignment processing on the multi-source image data, air quality data and occlusion data of the target street to obtain the multi-source feature dataset of the target street. The road segmentation module 302 is used to segment the target street based on the multi-source feature dataset to obtain a first road segment and a second road segment of the target street; the multi-source feature dataset of the first road segment includes street view images, remote sensing images, occlusion contour data and air quality data; the multi-source feature dataset of the second road segment includes remote sensing images, occlusion contour data and air quality data; The pixel-by-pixel semantic classification module 303 is used to perform pixel-by-pixel semantic classification on the street view image to obtain the street view visual index of the first road segment. The model prediction module 304 is used to input the remote sensing image of the second road segment into the trained deep learning regression model to obtain the street view visual index of the second road segment; the deep learning regression model is trained based on the street view visual index and remote sensing image of the first road segment. The data processing module 305 is used to perform visual analysis and spatial interpolation on the outline data of the obstruction and the air quality data respectively, and obtain the visual distance index and the air quality index respectively. The weighted fusion module 306 is used to perform weighted fusion of the street view visual index, the visible distance index and the air quality index to obtain the eye protection index. The eye protection index integration module 307 is used to integrate the eye protection index of the first road segment and the eye protection index of the second road segment to obtain the eye health assessment result of the target street.

[0089] Optionally, the pixel-by-pixel semantic classification module includes: The target street view image extraction unit is used to extract the target street view image that meets the human eye's panoramic vision conditions from the multi-source feature dataset of the first road segment; A pixel-by-pixel semantic classification unit is used to perform pixel-by-pixel semantic classification on the target street view image to obtain the semantic category of each effective pixel of the street view image; the semantic category includes vegetation and sky; The street view visual index determination unit is used to determine the green visibility index of the first road segment based on the effective pixel ratio of the vegetation; and to determine the sky visibility index of the first road segment based on the effective pixel ratio of the sky.

[0090] Optionally, the data processing module includes: A viewing angle determination unit is used to determine the viewing angle of a visible point on each segment of the target street. A visible direction distance determination unit is used to determine the visible direction distance from the visible point to the outline of the occluder based on the outline data of the occluder within the visible angle at the visible point. The visible distance index determination unit is used to obtain the visible distance index of each segment of the target street based on the visible direction distance.

[0091] Optionally, the data processing module further includes: The unit for determining the coordinates of the collection points for the air quality data is used to determine the coordinates of the collection points. The spatial distance determination unit is used to determine the spatial distance between the first road segment and the coordinates of the collection point when performing spatial interpolation on the air quality data of the first road segment. A semi-variogram determination unit is used to obtain the concentration of the target air pollutant in the air quality data, and the semi-variogram of the target air pollutant; An air quality index determination unit is used to determine the air quality index of the target air pollutant in the first road segment based on the spatial distance, the concentration, and the semi-variogram.

[0092] Optionally, the street eye health assessment system further includes: The remote sensing feature extraction module is used to extract features from the remote sensing image of the first road segment to obtain the remote sensing features of the first road segment. The dataset construction module is used to perform point-by-point pairing of the street view visual index and remote sensing features of the first road segment based on the positioning data of the first road segment, so as to obtain the training dataset and the validation dataset. The model training module is used to train a deep learning regression model based on the training dataset and the validation dataset; the trained deep learning regression model is used to predict the street view visual index of the second road segment.

[0093] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for assessing street-level eye health, characterized in that, include: Spatiotemporal alignment processing is performed on the multi-source image data, air quality data, and occlusion data of the target street to obtain the multi-source feature dataset of the target street. The target street is divided into road segments based on the multi-source feature dataset to obtain the first road segment and the second road segment of the target street; the multi-source feature dataset of the first road segment includes street view images, remote sensing images, occlusion outline data and air quality data; the multi-source feature dataset of the second road segment includes remote sensing images, occlusion outline data and air quality data. The street view image is subjected to pixel-by-pixel semantic classification to obtain the street view visual index of the first road segment; The remote sensing image of the second road segment is input into the trained deep learning regression model to obtain the street view visual index of the second road segment. The deep learning regression model was trained based on the street view visual index and remote sensing images of the first road segment. Visual analysis and spatial interpolation are performed on the outline data of the obstruction and the air quality data respectively to obtain the visual distance index and the air quality index respectively. The eye protection index is obtained by weighted fusion of the street view visual index, the visible distance index, and the air quality index. The eye protection index of the first road segment and the eye protection index of the second road segment are integrated to obtain the eye health assessment result of the target street.

2. The street eye health assessment method as described in claim 1, characterized in that, The step of performing pixel-by-pixel semantic classification on the street view image to obtain the street view visual index of the first road segment includes: Extract the target street view image that meets the human eye's panoramic vision condition from the multi-source feature dataset of the first road segment; The target street view image is subjected to pixel-by-pixel semantic classification to obtain the semantic category of each effective pixel in the street view image; the semantic category includes vegetation and sky. Based on the effective pixel ratio of the vegetation, the green visibility index of the first road segment is determined; based on the effective pixel ratio of the sky, the sky visibility index of the first road segment is determined.

3. The street eye health assessment method as described in claim 1, characterized in that, Visual analysis of the outline data of the obstruction yields a visual distance index, including: Determine the viewing angles of visible points on each segment of the target street; At the visible point, based on the outline data of the occluder within the visible angle, the visible directional distance from the visible point to the outline of the occluder is determined; Based on the visible direction distance, the visible distance index of each segment of the target street is obtained.

4. The street eye health assessment method as described in claim 1, characterized in that, Spatial interpolation of the air quality data yields the air quality index, including: Determine the coordinates of the data collection points for the air quality data; When performing spatial interpolation on the air quality data of the first road segment, the spatial distance between the first road segment and the coordinates of the collection point is determined; Obtain the concentration of the target air pollutant in the air quality data, and the semi-variogram of the target air pollutant; Based on the spatial distance, the concentration, and the semi-variogram, the quality index of the target air pollutant in the first road segment is determined.

5. The street eye health assessment method as described in claim 1, characterized in that, The step of performing pixel-by-pixel semantic classification on the street view image to obtain the street view visual index of the first road segment includes: Feature extraction is performed on the remote sensing image of the first road segment to obtain the remote sensing features of the first road segment; Based on the location data of the first road segment, the street view visual index and remote sensing features of the first road segment are paired point by point to obtain the training dataset and the validation dataset. A deep learning regression model is trained based on the training dataset and the validation dataset; the trained deep learning regression model is used to predict the street view visual index of the second road segment.

6. A street-based eye health assessment system, characterized in that, include: The spatiotemporal alignment module is used to perform spatiotemporal alignment processing on the multi-source image data, air quality data and occlusion data of the target street to obtain the multi-source feature dataset of the target street. The road segmentation module is used to segment the target street based on the multi-source feature dataset to obtain a first road segment and a second road segment of the target street; the multi-source feature dataset of the first road segment includes street view images, remote sensing images, occlusion contour data and air quality data; the multi-source feature dataset of the second road segment includes remote sensing images, occlusion contour data and air quality data; A pixel-by-pixel semantic classification module is used to perform pixel-by-pixel semantic classification on the street view image to obtain the street view visual index of the first road segment; The model prediction module is used to input the remote sensing image of the second road segment into the trained deep learning regression model to obtain the street view visual index of the second road segment; the deep learning regression model is trained based on the street view visual index and remote sensing image of the first road segment. The data processing module is used to perform visual analysis and spatial interpolation on the outline data of the obstruction and the air quality data respectively, and obtain the visual distance index and air quality index respectively. The weighted fusion module is used to weight and fuse the street view visual index, the visible distance index, and the air quality index to obtain the eye protection index; The eye protection index integration module is used to integrate the eye protection index of the first road segment and the eye protection index of the second road segment to obtain the eye health assessment result of the target street.

7. The street eye health assessment system as described in claim 6, characterized in that, The pixel-by-pixel semantic classification module includes: The target street view image extraction unit is used to extract the target street view image that meets the human eye's panoramic vision conditions from the multi-source feature dataset of the first road segment; A pixel-by-pixel semantic classification unit is used to perform pixel-by-pixel semantic classification on the target street view image to obtain the semantic category of each effective pixel of the street view image; the semantic category includes vegetation and sky; The street view visual index determination unit is used to determine the green view rate index of the first road segment based on the effective pixel ratio of the vegetation; and to determine the sky view index of the first road segment based on the effective pixel ratio of the sky.

8. The street eye health assessment system as described in claim 6, characterized in that, The data processing module includes: A viewing angle determination unit is used to determine the viewing angle of a visible point on each segment of the target street. A visible direction distance determination unit is used to determine the visible direction distance from the visible point to the outline of the occluder based on the outline data of the occluder within the visible angle at the visible point. The visible distance index determination unit is used to obtain the visible distance index of each segment of the target street based on the visible direction distance.

9. The street eye health assessment system as described in claim 6, characterized in that, The data processing module further includes: The unit for determining the coordinates of the collection points for the air quality data is used to determine the coordinates of the collection points. The spatial distance determination unit is used to determine the spatial distance between the first road segment and the coordinates of the collection point when performing spatial interpolation on the air quality data of the first road segment. A semi-variogram determination unit is used to obtain the concentration of the target air pollutant in the air quality data, and the semi-variogram of the target air pollutant; An air quality index determination unit is used to determine the air quality index of the target air pollutant in the first road segment based on the spatial distance, the concentration, and the semi-variogram.

10. The street eye health assessment system as described in claim 6, characterized in that, The street eye health assessment system also includes: The remote sensing feature extraction module is used to extract features from the remote sensing image of the first road segment to obtain the remote sensing features of the first road segment. The dataset construction module is used to perform point-by-point pairing of the street view visual index and remote sensing features of the first road segment based on the positioning data of the first road segment, so as to obtain the training dataset and the validation dataset. The model training module is used to train a deep learning regression model based on the training dataset and the validation dataset; the trained deep learning regression model is used to predict the street view visual index of the second road segment.

Images