High-resolution remote sensing image water body extraction method based on novel water body index

Abstract

The invention belongs to the technical field of remote sensing geoscience application, and particularly relates to a high-resolution remote sensing image water body extraction method based on a novel water body index. The method mainly comprises the following steps: 1) preprocessing a GF-2 remote sensing image to obtain the real reflectivity of the earth surface and improve the spatial resolution; 2) extracting a ground feature sample in the GF-2 remote sensing image, and analyzing the spectral characteristics of the ground feature sample; (3) a blue wave band, a green wave band, a red wave band and a near infrared (NIR) wave band are adopted, and an improved water body index (Near-Infringes-RGB WaterIndex, NRWI) is calculated; and 4) presetting a threshold value, and carrying out binarization segmentation on the NRWI index image to obtain a water body thematic map. The water body extraction method provided by the invention can inhibit background interference, so that rapid and effective extraction of the water body is realized, and a new method is provided for water body extraction.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing geoscience application technology, specifically relating to a method for extracting water bodies from high-resolution remote sensing images based on a novel water body index. Background Technology

[0002] Surface water, as a crucial component of terrestrial ecosystems, significantly impacts ecological sustainability and socio-economic development. Accurate identification and monitoring of the spatiotemporal dynamics of surface water are essential for climate monitoring, drought relief, flood control, and water resource allocation. Remote sensing technology, an emerging observational technique based on spectral theory for rapidly acquiring surface information, offers advantages such as large-area synchronous observation, periodic dynamic imaging, and independence from terrain limitations, making it highly suitable for comprehensive surface water identification across diverse landscapes.

[0003] Current water extraction methods are mainly divided into image classification methods and spectral index methods. Image classification methods treat water extraction as an image classification or segmentation task. Early research used a "feature engineering + machine learning classifier" paradigm, while current methods employ end-to-end training models based on deep learning. Spectral index methods use spectral features from different bands for band operations and extract water bodies by selecting thresholds. They are mostly used for remote sensing images with medium to low spatial resolution.

[0004] While image classification methods have high accuracy, they rely on large labeled datasets and computational resources, making them difficult to apply to real-time monitoring scenarios such as flood emergencies. Spectral feature analysis technology is computationally efficient, but it still faces challenges in distinguishing water bodies from urban features with similar spectra (including building shadows and dark, impermeable surfaces).

[0005] Traditional water body index calculations rely on the shortwave infrared (SWIR) band, but high-resolution satellites (such as WorldView and GF-2) lack SWIR bands. Although Normalized Water Indices (NDWI), High-Resolution Water Indices (HRWI), and Urban Water Indices (UWI) have been proposed for high-resolution data, the spectral obfuscation problem between water and non-water bodies in complex environments remains unresolved. These limitations highlight the shortcomings of high-resolution remote sensing imagery for water body extraction.

[0006] In view of this, the inventors aim to design a high-resolution remote sensing image water extraction method based on a novel water index. Summary of the Invention

[0007] The purpose of this invention is to solve the problems of band deficiency and spectral confusion in the water body extraction process of high-resolution remote sensing images, and to provide a water body extraction method based on a new water body index for high-resolution remote sensing images. This method can achieve efficient and accurate water body extraction from high-resolution remote sensing images, and provide effective assistance for flood monitoring, disaster management and urban water resource management.

[0008] To achieve the above-mentioned technical objectives and effects, the present invention is implemented through the following technical solution:

[0009] This invention provides a method for extracting water bodies from high-resolution remote sensing images based on a novel water index, comprising the following steps:

[0010] S1. Perform radiometric calibration and atmospheric correction on high-resolution remote sensing images to obtain the true surface reflectance;

[0011] S2. Perform panchromatic and multispectral fusion on high-resolution remote sensing images to generate multispectral images;

[0012] S3. Calculate the NRWI index value for each pixel to obtain the NRWI index image;

[0013] S4. Binarize the NRWI index image based on a preset threshold. Pixels with NRWI>=Threshold are water bodies; pixels with NRWI<=Threshold are non-water bodies.

[0014] Furthermore, in step S1, the high-resolution remote sensing image is either a high-resolution GF-2 image or a medium-resolution Sentinel-2 image.

[0015] Further, in step S1, for the radiometric calibration of the GF-2 high-resolution remote sensing image, the original digital quantization value is converted into the radiance at the sensor entrance pupil using the following formula:

[0016]

[0017] Where DN is the original digital quantization value; L λ The radiance of band λ is expressed in W·m. -2 ·sr -1 ·μm -1 Gain and Offset are the preset gain and offset parameters for the GF-2 satellite.

[0018] Furthermore, in step S1, atmospheric correction is performed using the FLAASH module of the ENVI platform to obtain the true surface reflectance ρ. λ :

[0019]

[0020] In the formula, d is the Earth-Sun distance correction factor, and E sun,λ Let λ be the solar irradiance in band λ, and θ be the solar zenith angle.

[0021] Furthermore, in step S2, the spatial resolution of the multispectral image is 1m.

[0022] Furthermore, in step S3, the formula for calculating the NRWI index value is as follows:

[0023]

[0024] In the formula, ρ green ρ is the reflectivity in the green light band. blue ρ is the reflectivity in the blue light band. NIR ρ is the reflectance in the near-infrared band. red This refers to the reflectivity in the red light band.

[0025] Furthermore, in step S4, for high-resolution GF-2 images, the preset threshold is set to 0.4 to 0.6.

[0026] Furthermore, for high-resolution GF-2 images, the preset threshold is set to 0.5.

[0027] Furthermore, in step S4, for the medium-resolution Sentinel-2 image, the preset threshold is set to 0 to 0.2.

[0028] Furthermore, for medium-resolution Sentinel-2 images, the preset threshold is set to 0.1.

[0029] The beneficial effects of this invention are:

[0030] 1. The water extraction method provided by this invention can achieve efficient and accurate extraction of water from high-resolution remote sensing images based on the NRWI index, marking an important step forward in water extraction under complex environments.

[0031] 2. The water extraction method provided by this invention is applicable to the differentiation of water bodies from building shadows and bright roofs in complex urban environments, and supports water extraction across sensors (GF-2 and Sentinel-2).

[0032] Of course, any product implementing this invention does not necessarily need to achieve all of the above advantages at the same time. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a schematic flowchart of the method of the present invention;

[0035] Figure 2 The study points for verifying the location and accuracy of the sample study area, along with corresponding GF-2 images and schematic diagrams of water samples;

[0036] Figure 3 The spectral reflectance curves of water and non-water pixels in the sample study area;

[0037] Figure 4 Box plots showing the pixel reflectance distribution of various land features;

[0038] Figure 5 Visual assessment maps of four water quality indices (NDWI, HRWI, UWI, NRWI) in different study locations;

[0039] Figure 6 This is a schematic diagram showing the accuracy evaluation results of four water body indices based on different thresholds in GF-2 imagery at various study locations.

[0040] Figure 7 This is a schematic diagram of the water body extraction results based on the optimal threshold for each water body index in GF-2 imagery.

[0041] Figure 8 A comparison chart of water extraction results between NRWI and deep semantic segmentation models in different scenarios;

[0042] Figure 9 This is a schematic diagram showing the accuracy evaluation results of four water indices based on different thresholds in Sentinel-2 images at various study locations.

[0043] Figure 10 The results of water body extraction based on the optimal threshold for each water body index in Sentinel-2 image. Detailed Implementation

[0044] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] like Figures 1-4 As shown, this embodiment provides a method for extracting water bodies from high-resolution remote sensing images based on a novel water index, including the following steps:

[0046] Step S1: Data Preprocessing

[0047] Radiometric calibration and atmospheric correction were performed on GF-2 remote sensing images to obtain the true surface reflectance; panchromatic and multispectral images were fused to obtain high spatial resolution GF-2 image data.

[0048] 1) Radiometric calibration: For radiometric calibration of GF-2 high-resolution remote sensing imagery, the following formula is used to convert the raw digital quantization values ​​into radiance at the sensor's entrance pupil:

[0049]

[0050] Where DN is the original digital quantization value; L λ The radiance of band λ is expressed in W·m. -2 ·sr -1 ·μm -1 Gain and Offset are the preset gain and offset parameters for the GF-2 satellite.

[0051] 2) Atmospheric correction: Atmospheric correction was performed using the FLAASH module of the ENVI platform to obtain the true surface reflectance ρ. λ :

[0052]

[0053] In the formula, d is the Earth-Sun distance correction factor, and E sun,λ Let λ be the solar irradiance in band λ, and θ be the solar zenith angle.

[0054] 3) Image fusion: The Gram-Schmidt PanSharpening algorithm is used to improve the spatial resolution of multispectral images to the panchromatic band level (GF-2 panchromatic band resolution is 1m). The specific process is as follows:

[0055] ①Band registration: Resample the multispectral image to panchromatic band resolution to ensure spatial alignment.

[0056] ② Gram-Schmidt transform: transform the panchromatic band P high With P low Perform orthogonal decomposition to extract spatial detail information.

[0057] ③ Detail Injection: The high-frequency detail components after decomposition are injected into the multispectral bands to generate a fused high-resolution multispectral image.

[0058] Step S2: Selection of Study Area

[0059] Artificial digital water body boundaries were constructed by combining GF-2 and Google Earth imagery, and 16 study areas were selected.

[0060] Step S3: Construction of Water Body Indices

[0061] 2400 pure pixels were selected to analyze the spectral characteristics of water bodies and non-water bodies, determine the differences in key bands, and construct a water body index by gradually optimizing the band combination and weights and introducing NIR bands. The calculation formula is as follows:

[0062]

[0063] In the formula, ρ green ρ is the reflectivity in the green light band. blue ρ is the reflectivity in the blue light band. NIR ρ is the reflectance in the near-infrared band. red This refers to the reflectivity in the red light band.

[0064] Step S4, Threshold Segmentation

[0065] Set an appropriate threshold (Threshold) to perform binarization segmentation on the NRWI index image. Pixels with NRWI >= Threshold are water bodies; pixels with NRWI <= Threshold are non-water bodies (typically, the preset optimal threshold for GF-2 image is 0.5, and the preset optimal threshold for Sentinel-2 image is 0.1).

[0066] 1) Threshold determination: The Otsu algorithm is used to automatically calculate the threshold that maximizes the inter-class variance. The formula is:

[0067] σ 2 (T)=w0(T)w1(T)μ0(T)-μ1(T) 2

[0068] Where w0 and w1 are the proportions of water body pixels and non-water body pixels, and μ0 and μ1 are the mean values.

[0069] 2) Binarization segmentation operation: Threshold segmentation is used to perform pixel-by-pixel segmentation of the NRWI index image. The formula is as follows:

[0070]

[0071] A specific application of this embodiment, based on the construction process of the NRWI index, is described below:

[0072] 1) Selection of study area

[0073] Sixteen study areas were selected in Anhui Province, covering various water body types (rivers, lakes, ponds, etc.) and complex terrain. GF-2 images of each area were radiometrically calibrated, atmospherically corrected, and fused to obtain 1m spatial resolution remote sensing images.

[0074] 2) Spectral Feature Analysis

[0075] Ten types of land features were selected from the study area, including four types of water bodies (green water, dark green water, cyan water, and black water) and six types of non-water bodies (building shadows, bright buildings, blue-roofed buildings, dark buildings, bare land, and vegetation). 2400 pure pixels (200 pixels for each of the ten land features) were extracted and their spectral characteristics were analyzed.

[0076] In the spectral characteristics of water bodies, due to the presence of phytoplankton and other suspended matter, green water bodies have high reflectance in the green band, while the reflectance in the red and near-infrared bands is low. Due to the high concentration of organic matter and the abundance of phytoplankton, the reflectance of dark green water bodies in the blue and green bands is lower than that of green water bodies, and even lower in the near-infrared band. Black water bodies have a high content of dissolved organic carbon, resulting in extremely low reflectance across the entire spectrum. Cyan water bodies are affected by specific algae or water quality parameters, and the reflectance peaks in the blue and green bands are significant.

[0077] In the spectral characteristics of non-aquatic bodies, the green band reflectance is higher than the near-infrared band reflectance. For building shadows, the reflectance is low across the entire spectrum, but the relative difference between the green and near-infrared bands makes their NDWI values ​​similar to those of water bodies. For bright buildings, the visible light band reflectance is high, and the ratio of the green to near-infrared bands results in NDWI > 0, overlapping with the NDWI of water bodies. For blue-roofed buildings, the blue and green band reflectances are high, and their spectral characteristics are confused with those of water bodies. The NDWI values ​​of the above-mentioned features significantly overlap with those of water bodies, making it difficult to distinguish between water and non-water bodies relying solely on NDWI.

[0078] 3) Construction of water body index

[0079] Based on the above findings, a new water quality index (NRWI) is proposed, defined as follows:

[0080]

[0081] In the formula, ρ green ρ is the reflectivity in the green light band. blue ρ is the reflectivity in the blue light band. NIR ρ is the reflectance in the near-infrared band. red This refers to the reflectivity in the red light band.

[0082] To better illustrate the technical effects of this invention, calculations were performed using the NDWI, HRWI, UWI, and NRWI indices for each research area. The resulting images are shown below. Figure 5As shown in the figure, while NDWI can identify water bodies in some areas, the NDWI values ​​of buildings and shadows in urban scenes highly overlap with those of water bodies, resulting in low discrimination accuracy. HRWI and UWI can detect water bodies, but the index values ​​of artificial surfaces and shadows are similar (both appearing bright yellow), making effective separation difficult. In contrast, the newly proposed NRWI shows significant advantages in all scenarios: the NRWI value of water bodies is significantly higher than that of non-water bodies, with clear boundaries. Even in complex urban or natural landscapes, it can accurately extract water body information, greatly reducing false positives and demonstrating its high discriminative power and robustness in spectrally confused scenarios.

[0083] 4) Threshold segmentation

[0084] The thresholds were selected based on statistical charts of the accuracy of water extraction in each study area at different thresholds. For example... Figures 6-7 As shown, the water body extraction performance of various water body indices at the optimal threshold was compared, highlighting the comprehensive advantages of NRWI: In complex urban scenes, NDWI and HRWI suffered from severe contamination due to building interference, while UWI, although partially suppressing artificial surfaces, still had misclassifications. NRWI, however, accurately separated water bodies from built-up areas through optimized band combinations (as shown in the example in the fourth row). In some areas (such as study area 1), NRWI and UWI successfully captured black water bodies that NDWI and HRWI missed (marked with green boxes). Furthermore, in simple rural scenes with less spectral interference (such as study area 6), the performance of each index was similar, but NRWI showed stronger background suppression capabilities in areas with complex land cover, effectively distinguishing water bodies from interfering features such as shadows and buildings, verifying its robustness and practicality in highly heterogeneous environments.

[0085] 5) Accuracy assessment

[0086] To comprehensively evaluate the water extraction performance of NRWI, mainstream deep learning models (UNet, DeepLabv3+) were selected for comparison. NRWI can extract water without any training samples, while deep learning models still require training samples; for example... Figure 8 As shown, NRWI performs similarly to deep learning models in most areas, but it misses some water samples (marked with white ellipses), while performing better in complex details (marked with black ellipses). With its advantages of not requiring sample training and high computational efficiency, NRWI demonstrates significant competitiveness in real-time monitoring and other practical applications, and is particularly suitable for large-scale water extraction tasks with limited resources or requiring rapid response.

[0087] To verify the cross-sensor adaptability of NRWI, water body extraction was performed using Sentinel-2 imagery to calculate NDWI, MNDWI, AWEI_sh, and NRWI indices. Figure 9As shown, NRWI achieved the highest water extraction accuracy in the 10 study areas, with the optimal threshold remaining stable in the 0-0.1 range. Figure 10 As shown, in complex urban scenes, NRWI can accurately separate water bodies from interfering features such as buildings and shadows. In natural landscapes, although its performance is similar to other indices (NDWI, MNDWI, etc.), its anti-aliasing ability is superior. The results indicate that, despite the need to adjust the threshold due to sensor characteristics, NRWI's cross-resolution stability and environmental adaptability validate its potential as a general-purpose water body index, suitable for rapid water body extraction from multi-source remote sensing data.

[0088] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A water body extraction method based on a new water body index for high-resolution remote sensing images, characterized in that, The method comprises the following steps: S1, radiometric calibration and atmospheric correction are performed on high-resolution remote sensing images to obtain real surface reflectivity; S2, panchromatic and multispectral band fusion is performed on the high-resolution remote sensing images to generate a multispectral image; S3, the NRWI index value of each pixel is calculated to obtain an NRWI index image; S4, the NRWI index image is binarized and segmented based on a preset threshold, and the pixels with NRWI >= Threshold are water bodies, and the pixels with NRWI <= Threshold are non-water bodies. 2.The method of claim 1, wherein, In step S1, the high-resolution remote sensing images are high-resolution GF-2 images or medium-resolution Sentinel-2 images. 3.The method of claim 2, wherein, In step S1, for the radiometric calibration of the GF-2 high-resolution remote sensing images, the following formula is used to convert the original digital quantization value into the radiance at the sensor entrance pupil: Wherein, DN is the original digital quantization value; L λ is the radiance of the wave band λ, unit W·m -2 ·sr -1 ·μm -1 ; Gain and Offset are the preset gain and offset parameters of GF-2 satellite. 4.The method of claim 3, wherein, In step S1, the atmospheric correction is performed by using the FLAASH module of the ENVI platform to obtain the surface real reflectance p λ : where d is the earth-sun distance correction factor, E sun,λ is the solar irradiance for wavelength band λ, and θ is the solar zenith angle. 5.The method of claim 1, wherein, In step S2, the spatial resolution of the multispectral image is 1 m. 6.The method of claim 1, wherein, In step S3, the calculation formula of the NRWI index value is as follows: In the formula, p green is the reflectivity in the green light band, p blue is the reflectivity in the blue light band, p NIR is the reflectivity in the near-infrared light band, and p red is the reflectivity in the red light band. 7.The method of claim 2, wherein, In step S4, for the high-resolution GF-2 images, the preset threshold is set to 0.4-0.

6. 8.The method of claim 2, wherein, In step S4, for the medium-resolution Sentinel-2 images, the preset threshold is set to 0-0.2.

Images