A method for converting far-field remote sensing images based on near-field images

Abstract

The present invention relates to a far-field remote sensing image conversion method based on a near-field image. The method comprises: performing pixel-level segmentation on a near-field high-resolution image to obtain the material category information and geometric structure information of the target area; establishing a ground object reflection model and using the ground object material category information and geometric structure information as input to obtain the ground end member reflectivity information; establishing a coupled nonlinear hybrid model of ground object reflection and atmospheric scattering, using the ground end member reflectivity information and solar radiation as input and combining with the remote sensing platform radiation calibration coefficient to obtain a far-field remote sensing image; at the same time, inverting the ground end member reflectivity based on the near-field camera radiation calibration coefficient and ground measured data, and obtaining a far-field remote sensing image through the above-mentioned coupled nonlinear hybrid model and the far-field platform radiation calibration coefficient, calibrating with the above-mentioned far-field image, and completing the conversion of the far-field remote sensing image. The present invention uses the near-field image and solar radiation as input to obtain a far-field remote sensing image. Compared with the existing remote sensing image simulation generation method, the present invention can realize a high-precision three-dimensional digital model without the need for a ground target area, and can also improve the quality and speed of image conversion generation.

Description

Technical Field

[0001] The present invention relates to the field of computer image processing, and in particular to a method for converting a far-field remote sensing image based on a near-field image. Background Art

[0002] With the continuous development of remote sensing technology, driven by military needs such as space reconnaissance, the spatial resolution of remote sensing sensors has been continuously improved. In addition, the flight and shooting of satellites are not restricted by national boundaries and geographical conditions, and more random and targeted shooting can be achieved. The application of aerospace satellite images is becoming more and more extensive, especially for military battlefield situation assessment and combat decision-making. In view of the limitations of satellite images such as long revisit time, difficulty in acquisition and high cost, high-quality near-field and far-field image conversion is of great significance to related research based on satellite images.

[0003] The existing far-field remote sensing image generation method is mainly generated by establishing a full-link simulation model. The full-link simulation model mainly includes ground three-dimensional digital simulation model, ground object reflection model, atmospheric transmission model, remote sensor imaging model, etc. By establishing a three-dimensional digital simulation model and giving geometry, material and other information, and then simulating and calculating the ground object reflection model, atmospheric transmission model, etc., the far-field remote sensing image is obtained, which is an important way to obtain satellite simulation images. However, there are the following limitations in practical applications: First, the huge three-dimensional simulation model has the problems of insufficient details, large amount of calculation, and slow simulation speed. Secondly, the ground object reflection model mostly uses a bidirectional reflection distribution function, but the measured spectral data is difficult to eliminate the scattering in the atmospheric downlink radiation process and the influence of multiple reflections of adjacent ground objects on the incident light, resulting in poor authenticity of the simulated image. The current research considers the ground effect and the atmospheric effect separately, which simplifies the problem, but does not consider the complex coupling effect of the ground effect and the atmospheric transmission process, which makes the research significance have certain limitations. Summary of the invention

[0004] The bidirectional reflectance distribution function can effectively describe the reflection characteristics of different ground objects. Based on the modeling of the bidirectional reflectance distribution function, the scattering effect of the atmospheric downlink radiation process and the reflection effect of nearby ground objects are fully considered for the incident radiation, and the ground object reflection model is optimized and corrected. At the same time, based on the image processing algorithm, with the help of pixel-level segmentation and inversion of near-field high-resolution images, the near-field image can be converted into a far-field remote sensing image without the need for a three-dimensional digital scene model.

[0005] (I) Technical Problems to be Solved by the Present Invention

[0006] The present invention provides a far-field remote sensing image conversion method based on near-field images. This method is an improvement and extension of the current remote sensing image simulation method, which fully considers the scattering effect and proximity effect of the atmospheric downward radiation process, optimizes the ground object reflection model of the bidirectional reflectance distribution function; considers the coupling of the mixed pixel model, and establishes a coupled nonlinear mixed model; and overcomes the current problems of poor simulation quality and slow speed by means of segmentation, inversion and calibration of near-field images.

[0007] (II) Technical solution adopted by the present invention

[0008] A method for converting a far-field remote sensing image based on a near-field image, characterized in that it comprises the following steps:

[0009] (1) Use image segmentation methods to perform pixel-level segmentation on near-field high-resolution images to obtain the material category information and geometric structure information of the target area;

[0010] (2) Comprehensively consider the atmospheric scattering effect and the reflection effect of nearby ground objects in the solar incident radiation process, establish a ground object reflectivity model, and use the ground object material category information and geometric structure information as input to calculate the ground end member reflectivity information;

[0011] (3) Establish a coupled nonlinear hybrid model from ground end-members to far-field pixels, calculate the radiance at the far-field entrance pupil based on the reflectance information of the ground end-members, and then convert the radiance into a far-field remote sensing image based on the radiation calibration coefficient of the far-field remote sensing platform;

[0012] (4) Based on the radiation calibration coefficient of the near-field camera platform and the measured reflectivity data on the ground, a mapping relationship between the near-field image and the measured data of the corresponding end-member on the ground is established, and the reflectivity information of the ground end-member is inverted. The far-field remote sensing image is obtained through the above-mentioned coupled nonlinear hybrid model and the far-field remote sensing radiation calibration coefficient, and is calibrated with the far-field remote sensing image obtained in step 3 to complete the conversion of the final far-field remote sensing image.

[0013] (III) Beneficial effects of the present invention

[0014] Due to the adoption of the above-mentioned technical scheme, the beneficial effects of the present invention are as follows: compared with the current remote sensing satellite simulation generation method, the model constructed by this method fully considers the comprehensive influence of the reflection characteristics of the ground objects, the atmospheric scattering effect, and the proximity effect of the ground objects, thereby improving the simulation conversion quality; making full use of the segmentation, inversion and calibration of the near-field image, it is possible to achieve the need to establish a ground three-dimensional digital simulation model, which can greatly save the conversion and generation time of the far-field remote sensing image; at the same time, the inversion mapping relationship of the near-field image has the universality for near-field images from different sources; the high-quality far-field remote sensing image generated by the conversion can be further applied to tasks such as target detection based on satellite images and battlefield damage assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without making any creative work.

[0016] Figure 1 : Flowchart of the far-field remote sensing image conversion method based on near-field images. DETAILED DESCRIPTION

[0017] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0018] (1) Perform pixel-level segmentation on near-field high-resolution images to obtain the material category information and geometric structure information of the target area;

[0019] A single pixel of a far-field remote sensing sensor covers a certain range of the surface when observing the earth. Due to the heterogeneous and isomorphic characteristics of surface objects, there may be objects of different material types or objects of the same material type but with different geometric structures within the field of view of a single pixel. Therefore, the detailed information on the material and geometric structure of objects in the target imaging area plays an important role in the quality of imaging.

[0020] This method establishes a semantic segmentation network model, trains and tests the near-field high-resolution image data set obtained from the target imaging area, and divides the image pixel values ​​by threshold according to the differences in the radiation illumination of the objects under different ground material categories and different geometric structures to achieve image segmentation. Finally, the pixels of the input image are divided into different categories of materials and different geometric structure features.

[0021] That is, the pixel-level segmentation method is constructed using a semantic segmentation network model to classify all pixel points of the input near-field image to obtain material information and geometric structure information of different categories of objects; among them, the geometric structure information is represented by pixel-level plane angle information and used for object reflectivity calculation to replace the establishment of large-scale high-precision three-dimensional digital models.

[0022] (2) Establishing a ground object reflectivity model, and using the ground object material category information and geometric structure information as input, calculating the ground end-member reflectivity information;

[0023] This method takes into account the precise expression of the ground object reflection model, fully considers the atmospheric scattering effect and the influence of the ground object proximity effect factors in the solar incident radiation process, and uses the integral form to optimize the single-direction incident spectrum of the bidirectional reflectance distribution function model into a ground object reflection model corresponding to a specific outgoing direction with multi-angle comprehensive incident in space; the ground object material information and geometric structure information obtained by the semantic segmentation model of the near-field image are used as input, and the ground object reflectivity information is output;

[0024] That is, the reflectivity model of the ground object is optimized based on the bidirectional reflectance distribution function, and the atmospheric scattering and multiple reflection effects of adjacent objects in the process of downward solar radiation are comprehensively considered. The incident radiation of the model is the comprehensive energy of the multi-angle incident radiation after multiple atmospheric scattering and reflection of adjacent objects.

[0025] (3) Establish a coupled nonlinear hybrid model from ground end-members to far-field pixels, calculate the radiance at the far-field entrance pupil based on the reflectance information of the ground end-members, and then convert the radiance into a far-field remote sensing image based on the radiation calibration coefficient of the far-field remote sensing platform;

[0026] This method uses the ground range corresponding to the near-field high-resolution image pixel as the endmember scale, integrates the linear synthesis of the heterogeneous and isomerous characteristics between the endmembers within the ground range corresponding to the remote sensing pixel and the nonlinear effects brought about by the proximity effect of the ground object and the atmospheric scattering, and establishes a coupled nonlinear hybrid model with strong correlation between the ground object and the atmosphere, so as to achieve high-quality conversion between the near-field pixel-level ground endmember data and the far-field low-resolution remote sensing pixel.

[0027] (4) Based on the radiation calibration coefficient of the near-field camera platform and the measured reflectivity data on the ground, a mapping relationship between the near-field image and the measured data of the corresponding end-member on the ground is established and the reflectivity information is inverted. The far-field remote sensing image is obtained through the above-mentioned coupled nonlinear hybrid model and the far-field remote sensing radiation calibration coefficient, and is calibrated with the above-mentioned far-field image to complete the conversion of the far-field remote sensing image.

[0028] With the near-field image and solar radiation as input, the far-field remote sensing image is inverted through the established mapping relationship and atmospheric transmission process, and the obtained image is calibrated with the remote sensing image obtained in step three, and finally the required far-field remote sensing image is output, thus completing the conversion process from near-field high-resolution image to far-field low-resolution remote sensing image.

[0029] It should be recognized that the above description is only a specific embodiment of the present invention, and the present invention is not limited to the specific structure shown or described above. The claims will cover all changes within the spirit and scope of the present invention.

Claims

1. A method for converting far-field remote sensing images based on near-field images, characterized in that The steps include: Step 1: Use image segmentation methods to perform pixel-level segmentation on the near-field high-resolution image to obtain the material category information and geometric structure information of the target area; Step 2: Establish a ground object reflectivity model, and use the ground object material category information and geometric structure information as input to calculate the ground end-member reflectivity information; Step 3: Establish a coupled nonlinear hybrid model from ground end-members to far-field pixels, calculate the radiance at the far-field entrance pupil based on the reflectance information of the ground end-members, and then convert the radiance into a far-field remote sensing image based on the radiation calibration coefficient of the far-field remote sensing platform; Step 4: Based on the radiation calibration coefficient of the near-field camera platform and the measured reflectivity data on the ground, a mapping relationship between the near-field image and the measured data of the corresponding end-member on the ground is established, and the reflectivity information of the ground end-member is inverted. The far-field remote sensing image is obtained through the above-mentioned coupled nonlinear hybrid model and the far-field remote sensing radiation calibration coefficient, and is calibrated with the far-field remote sensing image obtained in step 3 to complete the conversion of the final far-field remote sensing image; The pixel-level segmentation method in step 1 is constructed by using a semantic segmentation network model to classify all pixel points of the input near-field image to obtain different types of ground object material information and geometric structure information; wherein the geometric structure information is represented by pixel-level plane angle information and is used for ground object reflectivity calculation to replace the establishment of a large-scale high-precision three-dimensional digital model; The reflectivity model of the ground object established in the step 2 is optimized based on the bidirectional reflectance distribution function, taking into account the atmospheric scattering and multiple reflection effects of adjacent ground objects during the downward radiation of the sun. The incident radiation of the model is a combination of the multi-angle incident radiation energy after multiple atmospheric scattering and reflection of adjacent ground objects; In the step 3, the range of the ground object corresponding to the near-field high-resolution pixel is used as the ground object end member scale, and the heterogeneous characteristics of foreign objects, proximity reflection and atmospheric scattering effects between the end members are comprehensively considered to establish a coupled nonlinear hybrid model of ground object reflection and atmospheric scattering, and the multi-end data within the ground object range corresponding to the far-field remote sensing pixel is synthesized through proximity effect and atmospheric scattering to obtain the radiance at the far-field remote sensing entrance pupil, and the far-field remote sensing image is obtained by conversion through the remote sensing platform radiation calibration coefficient; The far-field remote sensing image obtained by inverting the mapping relationship between the near-field image and the ground measured data in step 4 is calibrated with the far-field remote sensing image obtained by near-field image segmentation, ground object reflection model, and coupled nonlinear hybrid model in step 3 to achieve high-quality conversion of the near-field image to the far-field image.

Images