CN-121981921-A - Shadow-removing ARD production method and device for high-resolution satellite image urban areas

CN121981921ACN 121981921 ACN121981921 ACN 121981921ACN-121981921-A

Abstract

The invention discloses a shadow removing ARD production method and device for a high-resolution satellite image urban area, and belongs to the technical field of remote sensing image processing. The method comprises the steps of screening out images covering urban areas, carrying out radiation calibration on high-resolution satellite images covering the urban areas to obtain apparent radiance images, carrying out preliminary shadow detection by adopting a pre-trained deep learning model based on the apparent radiance images, carrying out false detection elimination on a preliminary detection result through self-adaptive threshold segmentation by utilizing near infrared bands of earth surface reflectivity images after atmospheric correction to obtain a final shadow mask, carrying out histogram matching treatment on shadow areas on the earth surface reflectivity images after atmospheric correction based on the shadow mask to realize shadow removal, and finally outputting an ARD product with shadows removed. The invention obviously improves the shadow detection precision and the quantification reliability of ARD products, and is suitable for large-scale ARD production of urban high-resolution satellite images.

Inventors

CHEN XINGFENG
SUN YUAN
LI JIAGUO
ZHAO LIMIN
LI LI
LIU JUN
LI HUAFU

Assignees

中国科学院空天信息创新研究院

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. A method for producing shadow-removed ARD for a high-resolution satellite image urban area, comprising the steps of: Step 1, reading the center point coordinates of an original high-resolution satellite image, matching with preset urban vector data, and screening out image data covering an urban area; step 2, performing radiation calibration on the screened original high-resolution satellite images covering urban areas to obtain apparent radiance images; step 3, based on the apparent radiance image, performing preliminary shadow detection by adopting a pre-trained deep learning semantic segmentation model to obtain a preliminary shadow mask; Step 4, performing atmospheric correction on the apparent radiance image to obtain an earth surface reflectivity image, and performing false detection elimination on the preliminary shadow mask by utilizing a near infrared band in the earth surface reflectivity image to obtain a final shadow mask; And 5, on the surface reflectivity image, carrying out radiation recovery on the shadow area based on the final shadow mask, removing the shadow, and outputting analysis ready data ARD products.
2. The shadow-removing ARD production method for the high-resolution satellite image urban areas, according to claim 1, is characterized in that step 1 comprises the steps of loading a preset urban administrative division vector file, uniformly converting the vector file into a standard geographic coordinate system if a coordinate system of the vector file is inconsistent with a coordinate system of an original high-resolution satellite image central point, constructing a geometric object of the original high-resolution satellite image central point, judging whether the original high-resolution satellite image central point is contained in a polygonal geometric body of any one urban administrative division through a point-to-point inclusion relation algorithm, and outputting image data belonging to a covered urban area if the original high-resolution satellite image central point is contained.
3. The method according to claim 1, wherein the step 2 comprises automatically reading radiometric calibration parameters from metadata files of the original high-resolution satellite images covering the urban area, and converting digital quantized values of the original high-resolution satellite images covering the urban area into apparent radiance.
4. The shadow-removing ARD production method for the high-resolution satellite image urban areas according to claim 3 is characterized by comprising the steps of extracting red, green and blue wave bands from the apparent radiance image to form a three-channel image, carrying out pixel value mapping on the three-channel image, wherein pixel values in a 0% to 90% split range are linearly mapped to an integer interval of 0-255, setting pixel values above 90% split to 255 to obtain an input image in an 8-bit RGB format, inputting the input image to a pre-trained DeepLabv3+ semantic segmentation model, outputting a probability map of each pixel belonging to a shadow class by the DeepLabv3+ semantic segmentation model, and carrying out binarization processing on the probability map to obtain a preliminary shadow mask, wherein the pixel value corresponding to the shadow class is 1, and the pixel value corresponding to the non-shadow class is 0.
5. The method for shadow-removing ARD production of high-resolution satellite image urban areas according to claim 1, wherein in the step 4, false detection and elimination are performed on the preliminary shadow masks by utilizing near infrared bands in the surface reflectivity images, specifically, the reflectivity values of all pixels corresponding to the preliminary shadow mask areas in the near infrared bands are extracted, an optimal segmentation threshold value is automatically calculated by adopting an OTSU algorithm based on histogram distribution of the reflectivity values, and pixels with the reflectivity value higher than the optimal segmentation threshold value in the preliminary shadow mask areas are judged to be false detection ground objects and are eliminated from the preliminary shadow masks.
6. The method according to claim 1, wherein in step 5, the radiation recovery is achieved by histogram matching, and the histogram matching is to directly process the surface reflectivity image in the floating point data domain, and only the values of the pixels covered by the final shadow mask are adjusted, and the reflectivity values of the non-shadow regions remain unchanged.
7. The method for shadow-removing ARD production of high resolution satellite image urban areas according to claim 6, wherein the histogram matching method specifically comprises the steps of firstly performing morphological expansion operation on the final shadow mask to extract a non-shadow area at the periphery of a shadow boundary as a reference area for histogram matching, respectively counting histograms of the shadow area and the reference area on each wave band of the surface reflectivity image, automatically determining the optimal interval number of the histograms of each wave band based on a Freedman-Diaconis rule, and completing radiation recovery of the shadow area by mapping a cumulative distribution function of the shadow area to a cumulative distribution function of the reference area.
8. Shadow-removing ARD apparatus for producing in high-resolution satellite image urban area, characterized by comprising: The urban area judging module is used for reading the central point coordinates of the original high-resolution satellite images, matching the central point coordinates with preset urban vector data and screening out image data covering the urban area; The radiation calibration module is used for carrying out radiation calibration on the screened original high-resolution satellite images covering the urban area to obtain apparent radiance images; the deep learning shadow detection module is used for carrying out preliminary shadow detection by adopting a pre-trained deep learning semantic segmentation model based on the apparent radiance image to obtain a preliminary shadow mask; the atmosphere correction module is used for performing atmosphere correction on the apparent radiance image to obtain an earth surface reflectivity image; the near infrared false detection elimination module is used for eliminating false detection of the preliminary shadow mask by utilizing a near infrared band in the surface reflectivity image to obtain a final shadow mask; The shadow removing module is used for carrying out radiation recovery on a shadow area on the surface reflectivity image based on the final shadow mask so as to realize shadow removal; and the shadow removing ARD output module is used for outputting analysis ready data ARD products.
9. An electronic device, comprising: One or more processors; A memory for storing one or more programs; Wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the shadow-removing ARD production method of the high resolution satellite image urban area of any of claims 1-7.
10. A computer readable storage medium having stored thereon executable instructions which when executed by a processor cause the processor to implement the shadow-removing ARD production method of the high resolution satellite image urban area of any of claims 1-7.

Description

Shadow-removing ARD production method and device for high-resolution satellite image urban areas Technical Field The invention belongs to the technical field of remote sensing image processing, and particularly relates to a shadow removing ARD production method and device for a high-resolution satellite image urban area. Background The analysis ready data (ANALYSIS READY DATA, ARD) refers to remote sensing data products subjected to a series of standardized pre-processes, and aims to reduce the data processing burden of users and improve the data application efficiency. In urban high-resolution (high-resolution) satellite images, because of dense buildings and complex structures, the ARD quality is seriously affected by large-area shadows formed after sunlight is blocked. The shadows not only can mask important ground details such as roads, greenbelts, short facilities and the like around the building, so that the outline of the building is extracted incompletely and the ground classification is wrong, but also can interfere city quantitative analysis tasks such as building height inversion and volume rate statistics, and further increase the difficulty of application such as target identification and vegetation coverage detection. Therefore, effective shadow removal is a key technical link for improving the availability of the ARD of the urban high-resolution satellite images. The effect of shadow removal is highly dependent on the accuracy of shadow detection. Existing shadow detection methods mainly include model-based methods and feature-based methods. The method is characterized in that the method is used for constructing an illumination model by depending on prior parameters such as a digital surface model, a solar azimuth angle and the like, the data requirements are high, the applicability is limited, and the method is more commonly used for extracting characteristic differences of shadow areas in the aspects of spectrum, brightness, texture and the like. In recent years, a deep learning-based method becomes a research hotspot by virtue of strong feature learning capability, but most models are based on 8-bit RGB natural image training, when the method is directly applied to multispectral satellite images, the problem of obvious domain deviation exists due to the difference of radiation characteristics, wave band ranges and dynamic ranges, the detection precision is reduced, and dark ground objects such as water bodies, dark vegetation, blue roofs and the like are easily misjudged as shadows. In terms of shadow removal, existing methods mainly include gray-scale correction, filter enhancement, and methods based on generating an countermeasure network. Although the shadow area information can be recovered to a certain extent, global adjustment is often carried out on the whole image, the radiation value of a non-shadow area can be changed to influence the accuracy of subsequent quantitative application, a part of methods need to be processed in an 8-bit integer domain, the physical meaning of remote sensing data is destroyed, and moreover, a method based on deep learning usually needs a large amount of paired data training, so that the method is difficult to be suitable for large-scale and multi-scene batch processing. More importantly, in the whole process of the high-resolution satellite image ARD production, shadow detection and removal are not independent links, and are carried out cooperatively with preprocessing steps such as radiometric calibration, atmospheric correction and the like. Shadows exhibit different spectral response characteristics at different processing stages (e.g., apparent radiance and surface reflectivity), and the processing order of each link directly affects the final effect. However, the existing research focuses on a single technical link, lacks the overall consideration of system association and time sequence optimization between shadow processing and preprocessing flow, and restricts the large-scale and automatic production of high-quality shadow-removing ARD products. Therefore, a full-automatic processing method capable of deeply fusing key technologies such as shadow detection, false detection elimination and shadow removal and defining the optimal time sequence relation between the key technologies and preprocessing links such as radiometric calibration and atmospheric correction is needed, so that the problems of insufficient precision, flow splitting and difficulty in batch application in the prior art are solved, and a complete technical scheme is provided for large-scale production of shadow-removing ARD of urban high-resolution satellite images. Disclosure of Invention In order to solve the technical problems, the invention provides a shadow removing ARD production method and device for high-resolution satellite image urban areas, which creatively adopts an image collaborative strategy before and after atmospheric correction to carry out shadow detection and