Water body remote sensing image solar flare identification method based on multispectral features and XGBoost

Abstract

The invention relates to a water body remote sensing image solar flare identification method based on multispectral features and XGBoost, and the method comprises the steps: extracting a water body from a remote sensing image, screening out a region affected by solar flare and a normal water body region, marking the regions, and constructing a corresponding label data set; constructing an input feature data set by taking the atmospheric top reflection data of all wavebands of samples in the label data set and the spectral index data of the multi-waveband combination as input features; constructing a training set in combination with the input feature data set and the label data set, performing model training based on an XGBoost model, and constructing a solar flare recognition model; and carrying out solar flare identification on the to-be-identified water body remote sensing image based on the solar flare identification model. By adopting the method disclosed by the invention, the limitation of solar flare identification in a complex environment by a traditional single-band threshold method can be overcome, the identification precision and efficiency are improved, and a reliable technical means is provided for solar flare identification of large-area long-time-sequence remote sensing data.

Description

Technical Field

[0001] The present invention belongs to the technical field of remote sensing and environmental monitoring, and specifically relates to a method for identifying solar glare in water body remote sensing images based on multispectral features and extreme gradient boosting tree (XGBoost). Background Art

[0002] Lakes (including artificial reservoirs), as a vital component of the terrestrial water cycle, are not only crucial for maintaining regional ecological balance but also serve as strategic freshwater resources for the sustainable development of human society. These natural water bodies not only carry out fundamental socioeconomic functions such as urban and rural drinking water supply, agricultural irrigation, industrial water use, and aquaculture, but also play an irreplaceable role in ecological services such as flood regulation, biodiversity conservation, and carbon and nitrogen biogeochemical cycling. Sunlight is reflected specularly by the water surface. Sensors directly receive this reflection, resulting in excessive brightness in remote sensing images and interfering with the water's inherent spectral reflectance characteristics. This phenomenon, known as "solar glare," has long been a challenge in the field of water color remote sensing. Therefore, research on solar glare identification is of great significance to water color remote sensing and water environment monitoring.

[0003] On the one hand, strong solar glare obscures the water's own reflective information, resulting in significant data loss in remote sensing water color monitoring. This is particularly true in dynamic water monitoring and environmental modeling. Long-term monitoring is difficult due to data loss, which in turn impacts ecological assessment and decision support. On the other hand, solar glare interferes with the water's own spectral characteristics, causing brightness anomalies and color distortion in remote sensing images. This reduces the accuracy of water color extraction and introduces errors into the inversion results, further impacting water quality monitoring, particularly in the estimation of key parameters such as chlorophyll concentration and suspended solids. Solar glare represents a significant waste of resources. However, a mature and effective solar glare correction algorithm is currently lacking. Currently, most solar glare identification methods rely on manual visual interpretation or a single threshold, making it difficult to accurately identify solar glare over long time series, over wide areas, and with large amounts of data. Visual interpretation requires extensive prior knowledge and requires scene-by-scene interpretation of remote sensing images, which is labor-intensive. In addition, due to differences in satellite observation geometry, water optical properties and water surface conditions, solar glare varies in strength, which limits the band threshold extraction method in large areas and makes it difficult to achieve operational application.

[0004] Therefore, carrying out solar glare identification work can effectively provide a basis for subsequent solar glare correction, thereby reducing the interference of solar glare, and providing more accurate and consistent remote sensing data support for water environment protection, resource management and ecological monitoring. It provides more accurate data support for global water resources management, environmental protection and climate adaptation, and provides important technical guarantees for achieving sustainable development goals and responding to major challenges such as the global water crisis and ecological degradation. It has important scientific and application value. Summary of the Invention

[0005] The purpose of this invention is to address the limitations of existing solar flare identification methods and provide a solar flare identification method for water remote sensing imagery based on multispectral features and XGBoost. This method can achieve more accurate solar flare identification in complex environments, and the identification results provide a basis for solar flare correction.

[0006] To achieve the above-mentioned purpose, the technical solutions adopted by the present invention are as follows: A method for identifying solar glare in water remote sensing images based on multispectral features and XGBoost, the method comprising: Extract water bodies from remote sensing images, screen out areas affected by solar glare and normal water areas, annotate them, and construct corresponding label datasets; The input feature dataset is constructed using the atmospheric top reflectance data of all bands of samples in the label dataset and the spectral index data of multi-band combinations as input features; A training set is constructed by combining the input feature dataset and the label dataset, and a model is trained based on the XGBoost model to construct a solar flare recognition model; Based on the solar glare recognition model, solar glare recognition of the remote sensing image of the water body to be identified is performed.

[0007] In some embodiments of the present invention, the spectral index data of the multi-band combination includes a solar glare normalized index SGI1 constructed based on the shortwave infrared band and the green light band, and a solar glare difference index SGI2 constructed based on the near infrared band, which is calculated as follows: SGI1: [ρ TOA ( l swir )-ρ TOA ( l green )] / [ρ TOA ( l swir )+ρ TOA ( l green )] SGI2: [ρ TOA ( lnir1 )-ρ TOA ( l nir2 )] / ( l nir1 - l nir2 ) Where, ρ TOA ( l )refer to l Band top-of-atmosphere reflectivity, l swir 、 l green 、 l nir1 、 l nir2 They refer to the shortwave infrared band, green light band, near infrared band 1, and near infrared band 2 respectively.

[0008] In some embodiments of the present invention, the remote sensing image is resampled to unify the spatial resolution of different spectral bands.

[0009] In some embodiments of the present invention, the HydroLAKES dataset is used to mask the land in the remote sensing image to extract the water area.

[0010] Furthermore, based on the HydroLAKES dataset, several pixels are shrunk inward to create a land mask, thereby reducing the influence of land pixels. Preferably, 4 to 6 pixels are shrunk.

[0011] In some embodiments of the present invention, when constructing a label dataset, sun glare areas and normal water bodies are screened through visual interpretation.

[0012] In some embodiments of the present invention, the samples of the training set include image data of different lakes and reservoirs.

[0013] In some embodiments of the present invention, the data source of the remote sensing image is the Sentinel-2 MSI satellite sensor.

[0014] In some embodiments of the present invention, the plurality of single-band top-of-atmosphere reflectance data are top-of-atmosphere reflectance data of 13 bands of an MSI satellite sensor; The spectral index data of the multi-band combination is a solar glare normalized index constructed based on the 2202nm and 560nm bands and a solar glare difference index constructed based on the 842nm and 865nm bands.

[0015] The algorithm principle of the present invention is as follows: The present invention first extracts spectral information that characterizes glare from the resampled imagery. During the feature extraction process, solar glare areas and normal water areas are manually screened through visual interpretation. In addition to considering the spectral characteristics of all fixed bands of satellite sensors, the present invention introduces a normalized solar glare index (SGI1) based on the shortwave infrared and green bands as input feature parameters, and constructs a solar glare difference index (SGI2) based on the near-infrared band. The combination of SGI1 and SGI2 is used as an additional feature parameter. These features not only help distinguish between solar glare and non-glare areas, but also further improve the model's generalization capabilities under different glare conditions.

[0016] This paper employed manual visual interpretation to construct training samples. Samples containing both solar flare and non-sun flare were annotated in different regions of the imagery. All spectral information from these samples was extracted, along with constructed spectral combinations. These samples provided realistic labeled data for subsequent model training, ensuring a diverse and representative training set. To avoid overfitting, the training set included samples from imagery data from diverse geographic environments, thereby improving the model's generalization and adaptability.

[0017] After the training set was constructed, the extreme gradient boosting tree algorithm was selected for model training. 70% of the sample data set was randomly selected for training, and 30% of the sample data was used for model accuracy verification. The model was then fine-tuned using cross-validation.

[0018] As can be seen from the technical solution of the present invention described above, the multispectral feature-based solar flare identification method of the present invention first acquires remote sensing data and performs resampling and ground mask preprocessing. A training set is constructed by fusing multi-band spectral features and two spectral combinations. Model training and optimization are performed based on the extreme gradient boosting tree algorithm, and finally a solar flare identification model is established. The trained model is then used to identify solar flare. The method of the present invention helps to overcome the limitations of traditional single-band threshold methods for solar flare identification in complex environments, improves identification accuracy and efficiency, and provides a reliable technical means for solar flare identification in large-area, long-term remote sensing data. It has broad application prospects in fields such as lake water quality monitoring and marine environmental monitoring.

[0019] It should be understood that all combinations of the foregoing concepts and the additional concepts described in more detail below, as long as such concepts are not mutually inconsistent, can be considered part of the inventive subject matter of this disclosure. In addition, all combinations of the claimed subject matter are considered part of the inventive subject matter of this disclosure.

[0020] The foregoing and other aspects, embodiments, and features of the present invention will be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the present invention, such as features and / or beneficial effects of the exemplary embodiments, will become apparent from the following description or through practice of specific embodiments according to the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are not intended to be drawn to scale. In the drawings, each identical or nearly identical component shown in various figures may be represented by the same reference numeral. For the sake of clarity, not every component is labeled in every figure. Embodiments of various aspects of the present invention will now be described by way of example and with reference to the accompanying drawings, in which: Figure 1 This is the technical roadmap for the solution shown in Example 1.

[0022] Figure 2 This is a characteristic diagram of the solar glare and water spectrum in Example 1.

[0023] Figure 3 This is the characteristic parameter weight diagram of the solar glare recognition algorithm in Example 1.

[0024] Figure 4 This is the result of identifying the sun glare of the target lake in Example 1. DETAILED DESCRIPTION

[0025] In order to better understand the technical content of the present invention, specific embodiments are given below in conjunction with the accompanying drawings.

[0026] Various aspects of the present invention are described in this disclosure with reference to the accompanying drawings, in which a number of illustrative embodiments are shown. The embodiments of the present disclosure are not necessarily intended to include all aspects of the present invention. It should be understood that the various concepts and embodiments introduced above, as well as those described in more detail below, can be implemented in any of many ways, and that the concepts and embodiments disclosed herein are not limited to any implementation. In addition, some aspects of the present disclosure may be used alone or in any appropriate combination with other aspects of the present disclosure.

[0027] Example 1 This embodiment takes certain lakes and reservoirs as examples to further describe the technical solution of the present invention.

[0028] The present invention's solar flare identification method, based on multispectral features, first acquires remote sensing data of lakes from several different geographical environments as a training dataset and performs preprocessing, such as band resampling. Then, through manual visual interpretation, samples containing solar flare and non-glare are annotated in different areas of the image. Spectral information from these samples is extracted to construct spectral combinations. Based on this information, a training set is constructed, and the extreme gradient boosting tree algorithm is used for model training and optimization, ultimately establishing a solar flare identification model. Finally, the trained model is used to identify solar flare in remote sensing images of other lakes.

[0029] As an exemplary description, the implementation of the above method is described in detail below with reference to the accompanying drawings.

[0030] The solar glare recognition method based on multispectral features of the present invention includes the following steps: 1) Obtaining remote sensing images of lakes. In this example, remote sensing images of Lake Ulungur, Lake Bosten, Lake Qinghai, and Lake Tianmu are used to construct a training set for model training. The remote sensing image data source is the Sentinel-2 MSI satellite sensor.

[0031] Following the conventional lake environment remote sensing processing process, lake remote sensing data preprocessing and quality control were carried out: To ensure that the model can fully explore the potential information of the image, the original image data was first resampled. The purpose of resampling is to unify the spatial resolution of different spectral bands so that they have a consistent scale. Specifically, for bands with lower spatial resolution (such as 60 meters), we resample them to 10 meters to match the data of higher resolution bands. After image resampling, the HydroLAKES dataset was used to mask the land in the remote sensing image to remove interference from non-target areas. During the masking process, the land mask was shrunk inward by 5 pixels based on the HydroLAKES dataset.

[0032] 2) Construct a training set and extract spectral information that can characterize glare characteristics from the resampled images. During the feature extraction process, the solar glare areas and normal water areas were first manually screened through visual interpretation. In addition to considering the spectral characteristics of all fixed bands of the Sentinel-2 MSI sensor, this invention adds two additional spectral indices as additional feature input parameters: a normalized solar glare index constructed based on the shortwave infrared band and the green band, and a solar glare difference index constructed based on the near-infrared band. The calculation methods are as follows: Normalized Solar Glare Index (SGI1): [ρ TOA (2202)-ρ TOA (560)] / [ρ TOA (2202)+ρ TOA (560)] Solar glare difference index (SGI2): [ρ TOA (865)-ρ TOA (842)] / (865-842).

[0033] These features not only help to distinguish between sun flare and non-flare areas, but also further improve the model's generalization ability under different flare conditions.

[0034] The top atmosphere reflectance data of 13 bands of MSI sensor: ρ TOA (443),ρ TOA (560),ρ TOA (665),ρ TOA (705), ρ TOA (740),ρ TOA (783),ρ TOA (842), ρ TOA (865),ρ TOA (945),ρ TOA (1377),ρ TOA (1614),ρ TOA (2202), and two multi-band combination indices (normalized solar flare index and solar flare difference index) are used as input features of the model.

[0035] This paper employed manual visual interpretation to construct training samples. Samples containing both sun flare and non-sun flare were annotated in different regions of the imagery. All spectral information within these samples was read, along with the constructed spectral combinations (i.e., the input features). These samples provided realistic labeled data for subsequent model training, ensuring a diverse and representative training set. To prevent model overfitting, the training set included samples from different seasons and lighting conditions, thereby enhancing the model's adaptability.

[0036] 3) After the training set is constructed, the extreme gradient boosting tree model is used for model training.

[0037] 70% of the sample data set was randomly selected for training, and 30% of the sample data was used for model accuracy verification. The model was tuned through the cross-validation method to obtain the optimal model as the solar flare recognition model.

[0038] 4) Identification of solar glare The images of the target lakes (lakes a to c) for solar glare identification were resampled to a uniform resolution. The solar glare identification model constructed in step 3) was used to identify solar glare in the lake remote sensing images. The solar glare identification results and their accuracy were plotted. Figure 4As shown, the recognition result has high precision.

[0039] The method of the present application plays a great role in improving the available number of remote sensing images and the accuracy of water color remote sensing inversion, improving the remote sensing monitoring capability of inland water bodies such as lakes and reservoirs, and strengthening the security and sustainable management of water resources.

[0040] The sun glint recognition results of different lakes are as shown in Table 1 and Table 2. Figure 4 and Table 1.

[0041] It can be seen that the XGBoost model trained in Example 1 has good recognition accuracy for different lakes and has universality.

[0042] Table 1. Precision of sun glint recognition results of different lakes

[0043] Although the present application has been disclosed with the above preferred embodiments, it is not intended to limit the present application. Those skilled in the art can make various modifications and improvements without departing from the spirit and scope of the present application. Therefore, the protection scope of the present application shall be subject to the definition of the claims.

Claims

1. A method for identifying solar glare in water remote sensing images based on multispectral features and XGBoost, characterized in that: The method comprises: Extract water bodies from remote sensing images, screen out areas affected by solar glare and normal water areas, annotate them, and construct corresponding label datasets; The input feature dataset is constructed using the atmospheric top reflectance data of all bands of samples in the label dataset and the spectral index data of multi-band combinations as input features; A training set is constructed by combining the input feature dataset and the label dataset, and a model is trained based on the XGBoost model to construct a solar flare recognition model; Based on the solar glare recognition model, solar glare recognition of the remote sensing image of the water body to be identified is performed.

2. The method according to claim 1, wherein The spectral index data of the multi-band combination includes the solar glare normalized index SGI1 constructed based on the shortwave infrared band and the green light band, and the solar glare difference index SGI2 constructed based on the near infrared band. The calculation method is as follows: SGI1: [p TOA ( λ swir )-r TOA ( λ green )] / [p TOA ( λ swir )+r TOA ( λ green )] SGI2: [p TOA ( λ nir1 )-r TOA ( λ nir2 )] / ( λ nir1 - λ nir2 ) Where, ρ TOA ( λ )refer to λ Band top-of-atmosphere reflectivity, λ swir 、 λ green 、 λ nir1 、 λ nir2 They refer to the shortwave infrared band, green light band, near infrared band 1, and near infrared band 2 respectively.

3. The method according to claim 1, characterized in that The remote sensing images are resampled to make the spatial resolution of different spectral bands uniform.

4. The method according to claim 1, wherein The HydroLAKES dataset is used to mask the land in remote sensing images and extract water areas.

5. The method according to claim 4, characterized in that Based on the HydroLAKES dataset, several pixels are shrunk inward to create a land mask.

6. The method according to claim 1, characterized in that When constructing the label dataset, sun glare areas and normal water bodies were screened through visual interpretation.

7. The method according to claim 1, characterized in that The samples of the training set include image data of different lakes and reservoirs.

8. The method according to claim 1 or 2, characterized in that The data source of the remote sensing image is the Sentinel-2 MSI satellite sensor.

9. The method according to claim 8, characterized in that The multiple single-band top atmosphere reflection data are top atmosphere reflection data of 13 bands of the MSI satellite sensor; The spectral index data of the multi-band combination is a solar glare normalized index constructed based on the 2202nm and 560nm bands and a solar glare difference index constructed based on the 842nm and 865nm bands.

Images