Inland water atmosphere correction method based on high-resolution satellite image

Abstract

The invention provides an inland water atmosphere correction method based on a high-resolution satellite image, and belongs to the field of remote sensing science. The problems that an existing atmospheric correction method for the inland water body is not mature enough and is poor in adaptability to different water qualities are solved. Comprising the following steps: preprocessing a high-resolution original satellite image, correcting the preprocessed image, and carrying out space-time fusion on the corrected image and an image of a Landsat reflectivity product; resampling the image of the Landsat reflectivity product to obtain an image with the same resolution as the fused image; training a neural network model by taking auxiliary data such as the fused image, geometric information, aerosol optical thickness and the like as input and taking the resampled image of the Landsat reflectivity product as output; and inputting the preprocessed and corrected high-temporal-spatial-resolution satellite image, geometric information, aerosol optical thickness and other auxiliary data into the trained neural network model for water atmosphere correction.

Description

Technical Field

[0001] The invention provides an inland water body atmosphere correction method based on high-resolution satellite images, belonging to the field of remote sensing science and technology. Background Art

[0002] There is atmospheric interference between the complex water surface and the high-altitude sensors. The reflectivity of water depends on the interaction between light and matter. Most of the radiation comes from atmospheric scattering, mainly including Rayleigh scattering, aerosol scattering and water surface foam reflection. Compared with vegetation, soil and buildings on land, water bodies are low-reflection features. Under the influence of atmospheric components, the water body signal received by satellite sensors at the top of the atmosphere is generally only 8%-10%. Therefore, it is necessary to calculate the atmospheric absorption and scattering effects, effectively remove atmospheric interference from the total radiation value received by the sensor, and perform atmospheric correction on the water body, which is the basic work of water color remote sensing.

[0003] The traditional water body atmospheric correction technology uses clean water bodies as dark pixels. By assuming that the water-leaving radiation in the infrared band is 0, the signal in the infrared band is all contributed by atmospheric scattering, and the aerosol scattering in other bands is inferred to achieve atmospheric correction. However, this method is relatively rough. For the more turbid Class II water bodies, the reflectivity in the infrared band is often not 0. Therefore, when this method is used for water body correction, the reflectivity often has negative values, that is, over-correction. In response to this situation, some methods have been improved, such as using nearby clean water bodies to invert AOD, assuming that the short-wave infrared band reflectivity is 0, or that the infrared band ratio has spatial uniformity, but these methods all have certain limitations.

[0004] Another method uses radiation transfer models to simulate the absorption and scattering process of the atmosphere. It mainly includes the 6S model, MODTRAN model, etc. However, the radiation transfer model method usually requires the input of a large number of synchronous measured parameters, such as aerosol type, visibility, etc., which are often difficult to obtain, and the traditional radiation transfer model is difficult to separate the information of the atmosphere and water from the spectral information; therefore, some scholars use deep learning models to construct a nonlinear relationship between the TOA radiation at the top of the atmosphere and the reflectivity of the water body to simulate the atmospheric correction process. This type of method still has great potential in the application of atmospheric correction of water bodies. Summary of the invention

[0005] At present, the atmospheric correction methods for inland water bodies are not mature enough, and most of them have strong limitations and weak adaptability to different water qualities. Therefore, the present invention proposes an atmospheric correction method for inland water bodies based on high-resolution satellite images, which reduces some uncertain factors in the inversion of inland water reflectance using high-resolution satellite images. The method does not rely on the assumptions of traditional methods, has good universality, and is suitable for atmospheric correction of inland water bodies using high-resolution satellite images.

[0006] The technical solution adopted by the present invention is: a method for atmospheric correction of inland water bodies based on high-resolution satellite images, comprising the following steps: A. Constructing the data set: preprocessing the high-resolution original satellite images, then correcting the preprocessed images, and spatiotemporally merging the corrected images with the images of Landsat reflectivity products to obtain intermediate images with high spatiotemporal resolution; Resample the Landsat reflectance product images to obtain images with the same resolution as the high temporal and spatial resolution intermediate images; B. Training model: Use high temporal and spatial resolution intermediate images and auxiliary data as input, and resampled Landsat reflectivity product images as output for simulation training of the neural network model; C. Atmospheric correction: The preprocessed and corrected high-temporal and spatial resolution satellite images and auxiliary data are input into the trained neural network model to perform water body atmospheric correction and output the water body reflectivity.

[0007] Furthermore, the preprocessing in step A includes image fusion, radiometric calibration and geometric correction.

[0008] Furthermore, the correction of the preprocessed image in step A specifically includes Rayleigh correction and gas transmittance correction.

[0009] Furthermore, the auxiliary data in steps B and C include geometric information of the image, aerosol optical depth, wind speed, chlorophyll concentration and suspended particle concentration.

[0010] Furthermore, in step A, a dataset is obtained by fusing the corrected images at different time points and spatial locations with the images of Landsat reflectance products, thereby obtaining a higher temporal resolution and a more comprehensive dataset.

[0011] The beneficial effects of the present invention compared with the prior art are as follows: 1. The deep learning model has powerful feature learning and representation capabilities. It can automatically extract useful features from raw data and learn the relationship between these features. The model is trained to identify the key influencing factors in the water reflectance signal and establish a mathematical model between them. The deep learning method is used to mine the relationship between the influencing factors of the water reflectance signal and obtain stable atmospheric correction results. This is a complex process. In this process, a variety of auxiliary data are used as input to provide rich information and context for the deep learning model, so that it can more accurately capture and model the relationship between the influencing factors. Use the deep learning model to process and analyze these multi-source data.

[0012] 2. By fusing data image sets at different time points and spatial locations, higher temporal resolution and more comprehensive data sets can be obtained. This enables the deep learning model to make more full use of the temporal variation information and spatial distribution characteristics in the data, thereby improving the accuracy and stability of atmospheric correction.

[0013] 3. Before atmospheric correction, Rayleigh correction and gas transmittance correction are required to eliminate the influence of atmospheric scattering and absorption on the water reflectance signal. These steps can preferentially remove relatively stable atmospheric image factors and reduce uncertainty in the deep learning training process. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below in conjunction with the accompanying drawings: Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0015] like Figure 1 As shown, the present invention provides an inland water body atmospheric correction method based on high-resolution satellite images, which mainly includes the following steps: A. Use high-resolution images and Landsat reflectivity data images to build a dataset: Preprocess high-resolution original satellite images, including image fusion, radiometric calibration, and geometric correction; Then the preprocessed images were subjected to Rayleigh correction and gas transmittance correction; The preliminary corrected images and Landsat reflectivity product data are temporally and spatially fused to obtain intermediate images (time series images) with high temporal and spatial resolution. High-resolution time series images can be added and used as the input for neural network training. The data of Landsat reflectivity products are resampled to obtain the same resolution as the fused high-temporal and spatial-resolution intermediate image, which is used as the output of the neural network training.

[0016] B. Training model and optimizing auxiliary parameters: Use high temporal and spatial resolution intermediate images, geometric information, aerosol optical depth (AOD), wind speed, chlorophyll concentration, and suspended particle concentration as input, and the resampled Landsat reflectance product data as output for neural network simulation training, of which 70% is used for training data and 30% is used for verification, quantify the simulation effect, and adjust the model parameters.

[0017] C. Atmospheric correction: Use the intermediate image with high temporal and spatial resolution and auxiliary data (AOD, geometric information, etc.) as input to perform atmospheric correction on water bodies and output the water body reflectance.

[0018] The present invention uses a simulation method of deep learning, takes a variety of auxiliary data as input, mines the relationship between various influencing factors in the water body reflectance signal, and obtains a stable atmospheric correction result. Through time and space fusion, a high-order data image set can be obtained for atmospheric correction, and a sufficient amount of training data can be obtained. Before performing atmospheric correction, preprocessing, Rayleigh correction, and gas transmittance correction are performed, and the more stable atmospheric image factors are preferentially removed to reduce the uncertainty of deep learning training.

[0019] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for atmospheric correction of inland water bodies based on high-resolution satellite images, characterized in that: The following steps are involved: A. Constructing the data set: preprocessing the high-resolution original satellite images, then correcting the preprocessed images, and spatiotemporally merging the corrected images with the images of Landsat reflectivity products to obtain intermediate images with high spatiotemporal resolution; Resample the Landsat reflectance product images to obtain images with the same resolution as the high temporal and spatial resolution intermediate images; B. Training model: Use high temporal and spatial resolution intermediate images and auxiliary data as input, and resampled Landsat reflectivity product images as output for simulation training of the neural network model; C. Atmospheric correction: The preprocessed and corrected high-temporal and spatial resolution satellite images and auxiliary data are input into the trained neural network model to perform water body atmospheric correction and output the water body reflectivity.

2. The inland water atmospheric correction method based on high-resolution satellite images according to claim 1, characterized in that: The preprocessing in step A includes image fusion, radiometric calibration and geometric correction.

3. The inland water atmospheric correction method based on high-resolution satellite images according to claim 2, characterized in that: The correction of the preprocessed image in step A specifically includes Rayleigh correction and gas transmittance correction.

4. The inland water atmospheric correction method based on high-resolution satellite images according to claim 1, characterized in that: The auxiliary data in steps B and C include image geometry, aerosol optical depth, wind speed, chlorophyll concentration, and suspended particle concentration.

5. The inland water atmospheric correction method based on high-resolution satellite images according to claim 1, characterized in that: In step A, a dataset is obtained by fusing the corrected images at different time points and spatial locations with the images of Landsat reflectance products, thereby obtaining a higher temporal resolution and a more comprehensive dataset.

Images