Search

CN-115909074-B - Urban vegetation high-resolution drawing method based on trans-scale vegetation index guidance

CN115909074BCN 115909074 BCN115909074 BCN 115909074BCN-115909074-B

Abstract

The invention provides a city vegetation high-resolution drawing method based on trans-scale vegetation index guidance. The method mainly comprises the steps of 1) carrying out preprocessing operations such as resampling, geometric registration, radiation correction and the like on Sentinel-2L 2A images and high-resolution RGB images which cover the same area in two views and are similar in acquisition time, 2) providing a new cross-scale vegetation index, preliminarily extracting urban vegetation coverage, automatically obtaining urban vegetation and non-urban vegetation samples through threshold segmentation, and 3) training a semantic segmentation model DeepLab v & lt+ & gt by utilizing the automatically generated samples, wherein the trained model can be directly predicted to obtain a high-resolution drawing result of the urban vegetation. The method has the technical advantages that no manual sample labeling is needed, a high-resolution drawing result of the urban vegetation is automatically obtained, and a solution with low cost and good effect can be provided for large-scale urban vegetation high-resolution drawing.

Inventors

  • LIN CONG
  • FU JUNHAO
  • ZHANG PENG
  • ZHOU MENGXIAO
  • XU JIAWEI
  • HU CHUNXIA

Assignees

  • 南京市测绘勘察研究院股份有限公司

Dates

Publication Date
20260512
Application Date
20221206

Claims (7)

  1. 1. The urban vegetation high-resolution mapping method based on the trans-scale vegetation index guidance is characterized by comprising the following steps of: S1, aiming at the same region, acquiring a Sentinel-2L2A image and a high-resolution RGB image with similar time, and resampling the Sentinel-2L2A image to ensure that the spatial resolution is consistent with the high-resolution RGB image; S2, aiming at the two images in the step S1, selecting a plurality of pixel points as control points, and performing spatial registration operation on the Sentinel-2L 2A image and the high-resolution RGB image to enable the two images to coincide in spatial position; S3, relative radiation correction operation is carried out on red wave bands of the high-resolution RGB images by taking the earth surface reflectivity of the Sentinel-2L 2A image as a reference, and the red wave band reflectivity red RGB of the original RGB images is corrected to red RGB ', so that the earth surface reflectivity difference distance of the two images is reduced; s4, calculating a trans-scale vegetation index CSVI (i,j) of each position point of the area based on the two images processed in the steps S1-S3 to obtain a CSVI image, wherein the trans-scale vegetation index CSVI (i,j) of each position point has a calculation formula as follows: Wherein CSVI (i,j) represents a cross-scale vegetation index of an ith row and a jth column in a CSVI image, green RGB(i,j) 、blue RGB(i,j) represents surface reflectivities of a green wave band and a blue wave band of the ith row and the jth column of the high-resolution RGB image respectively, NIR S2(i,j) represents surface reflectivities of a near infrared wave band of the ith row and the jth column of the Sentinel-2L2A image, red' RGB(i,j) represents surface reflectivities of the ith row and the jth column of the high-resolution RGB image after relative radiation correction, and a numerical value n is used for expanding a cross-scale vegetation index range; s5, setting a threshold T fg 、T bg , and generating a pseudo sample S of urban vegetation and non-urban vegetation through threshold segmentation CSVI images, wherein the specific method is as follows: Wherein S (i,j) represents a sample label of a pixel point in an ith row and a jth column in CSVI image, when the label value is 0, the pixel point is non-urban vegetation, when the label value is 1, the pixel point is urban vegetation, and when the label value is 2, the pixel point is of an uncertain type and is ignored; And S6, using the pseudo sample S generated in the step S5 to train a semantic segmentation model DeepLab v3+, and further predicting to obtain a high-resolution drawing result of the urban vegetation after model training is completed.
  2. 2. The method of high resolution mapping of urban vegetation based on trans-scale vegetation index guidance according to claim 1, wherein in step S1, the resampling operation employs a nearest neighbor method.
  3. 3. The method for mapping urban vegetation at high resolution based on the guidance of a cross-scale vegetation index according to claim 1, wherein in step S2, the spatial registration method is implemented by applying Georeferencing tools built in ArcGIS software and adopting a second order polynomial transformation method.
  4. 4. The method of high resolution mapping of urban vegetation based on trans-scale vegetation index guidance according to claim 1, wherein the number of selected control points in step S2 is in the range of 20 or more.
  5. 5. The method of high resolution mapping of urban vegetation based on trans-scale vegetation index guidance according to claim 1, wherein in step S5, the threshold T fg is set to 0.3 and the threshold T bg is set to 0.2.
  6. 6. A computer-readable storage medium storing a computer program, wherein the computer program causes a computer to perform the urban vegetation high resolution mapping method according to any of claims 1-5.
  7. 7. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the urban vegetation high resolution mapping method of any of claims 1-5 when the computer program is executed.

Description

Urban vegetation high-resolution drawing method based on trans-scale vegetation index guidance Technical Field The invention relates to the technical field of remote sensing measurement, in particular to a urban vegetation high-resolution drawing method based on trans-scale vegetation index guidance. Background With the rapid development of the urban process, urban population is rapidly increasing. Urban ecosystems are facing unprecedented challenges, including a range of environmental problems such as climate change, reduced biodiversity, and various pollution. More and more people consider that urban vegetation (urban vegetation, UV) has beneficial effects on relieving such environmental stress, such as carbon sequestration, noise reduction, alleviation of urban heat island effects, improvement of urban resident physical and mental health, and the like. Therefore, accurate and timely urban vegetation distribution information is important for urban planning and management and urban environment research. Because the remote sensing technology has the advantages of multi-scale, repeated observation, wide coverage and the like, the method is widely used for vegetation mapping of various scenes. Compared with natural vegetation, urban vegetation surfaces appear as smaller, more fragmented patches on the remote sensing image, and the background environment is more complex, which makes urban vegetation extraction more challenging. Multispectral images (multi-SPECTRAL IMAGE, MSI) with medium spatial resolution (e.g., landsat and Sentinel-2) are widely used to map and monitor vegetation. Because it contains rich spectral information, MSI can provide key spectral features of vegetation that are critical to urban vegetation extraction. The normalized vegetation index (normalized difference vegetation index, NDVI) is an effective index for extracting and monitoring vegetation coverage because it takes advantage of the most obvious vegetation characteristics, i.e., the high reflectance properties of the Near Infrared (NIR) band and the strong absorption properties of the red wavelength. Although MSI is suitable for a wide range of vegetation drawings, sufficient spatial detail cannot be captured due to its low spatial resolution. Accurate urban vegetation coverage and its change information are critical to urban management, and therefore it is necessary to develop a high resolution urban vegetation mapping method. In recent years, high-resolution RGB images have become the most commonly used data source for fine mapping of urban vegetation by virtue of their lower cost compared to multispectral sensors. The vegetation mapping method using RGB images can be classified into a vegetation mapping based on a spectral index and a vegetation mapping based on a classifier. Currently, there are NGRDI indexes of vegetation using only RGB bands [ see Tucker C J. Red and photographic infrared linear combinations for monitoring vegetation[J]. Remote sensing of Environment, 1979, 8(2): 127-150.]、GLI [, Louhaichi M, Borman M M, Johnson D E. Spatially located platform and aerial photography for documentation of grazing impacts on wheat[J]. Geocarto International, 2001, 16(1): 65-70.]、RGBVI [, Bendig J, Yu K, Aasen H, et al. Combining UAV-based plant height from crop surface models, visible, and near infrared vegetation indices for biomass monitoring in barley[J]. International Journal of Applied Earth Observation and Geoinformation, 2015, 39: 79-87.]、MGRVI [, Bendig J, Yu K, Aasen H, et al. Combining UAV-based plant height from crop surface models, visible, and near infrared vegetation indices for biomass monitoring in barley[J]. International Journal of Applied Earth Observation and Geoinformation, 2015, 39: 79-87.]、VARI [, Gitelson A A, Kaufman Y J, Stark R, et al. Novel algorithms for remote estimation of vegetation fraction[J]. Remote sensing of Environment, 2002, 80(1): 76-87.], etc., which are generally designed based on the characteristics that vegetation has a higher reflectivity in the green band and a lower reflectivity in the red and blue bands. However, studies have shown that the spectral information contained in the red, green and blue bands is too small to perform accurate vegetation mapping [ see R. Neyns and F. Canters, "Mapping of Urban Vegetation with High-Resolution Remote Sensing: A Review," Remote Sens., vol. 14, no. 4, 2022.].. Some non-vegetation fields (such as water bodies) have similar spectral characteristics to vegetation, namely, the reflectivity of the green band is high, and the reflectivity of the red and blue bands is low, so that the above indexes are difficult to accurately distinguish vegetation from other land covers. Some studies combine lidar data and RGB images to facilitate urban vegetation extraction [ see N. Audebert, B. Le Saux, and S. Lefèvre, "Beyond RGB: Very high resolution urban remote sensing with multimodal deep networks," ISPRS J. Photogramm. Remote Sens., vol. 140, pp. 20–32, 2018]