CN115797362A

CN115797362A - Quality evaluation method and device for high-resolution remote sensing image and electronic equipment

Info

Publication number: CN115797362A
Application number: CN202310102505.8A
Authority: CN
Inventors: 范磊; 田静国; 王宇翔; 陈大光; 黄非; 关元秀; 王硕; 殷慧; 苏鑫; 付丽荣
Original assignee: Aerospace Hongtu Information Technology Co Ltd
Current assignee: Aerospace Hongtu Information Technology Co Ltd
Priority date: 2023-02-13
Filing date: 2023-02-13
Publication date: 2023-03-14
Anticipated expiration: 2043-02-13
Also published as: CN115797362B

Abstract

The invention provides a quality evaluation method and device of a high-resolution remote sensing image and electronic equipment, wherein the quality evaluation method comprises the following steps: obtaining an original high-resolution remote sensing image to be evaluated; generating an apparent reflectivity image based on metadata corresponding to the original high-resolution remote sensing image; performing object-oriented image segmentation processing on the original high-resolution remote sensing image based on the apparent reflectivity image to obtain an object-level segmentation image; performing pixel-level quality evaluation on the original high-resolution remote sensing image based on the apparent reflectivity image and a preset decision tree to obtain a pixel-level quality evaluation image; and determining a target quality evaluation image corresponding to the original high-resolution remote sensing image according to the object-level segmentation image and the pixel-level quality evaluation image. The method can obviously improve the accuracy of quality evaluation of the high-resolution remote sensing image.

Description

Quality evaluation method and device for high-resolution remote sensing image and electronic equipment

Technical Field

The invention relates to the technical field of remote sensing image processing, in particular to a quality evaluation method and device for high-resolution remote sensing images and electronic equipment.

Background

At present, two methods for evaluating optical remote sensing images, which are mainstream and can be used for engineering application, are Fmask (Function of mask) algorithm facing a landsat image and SC (Scene Classification) algorithm facing a sentinel2 image, respectively, and the basic ideas of Fmask and SC algorithms are based on a remote sensing image atmospheric apparent Reflectance product (TOA, top of atmospheric Reflectance), so that areas of cloud, cloud shadow, snow, water body and the like of image quality are detected, and a pixel-level quality evaluation image product is generated. The method has the following problems in high resolution image engineering application: (1) The object-oriented segmentation technology is not utilized, and the precision is limited; (2) Usually, 400-11000nm waveband information is used, including visible light, near infrared, intermediate infrared and thermal infrared waveband information, and a high-resolution image only has 400-1000nm waveband information, namely only visible light and near infrared waveband information, so that the method is difficult to be applied to the high-resolution image; (3) Because the orbit of the high-score satellite is low, the mountain shadow caused by the fluctuation of the mountains is also an important interference factor for the image quality, and the method cannot detect the mountain shadow. In conclusion, the existing scheme has the problem of poor accuracy of optical remote sensing image evaluation.

Disclosure of Invention

In view of this, the present invention provides a quality evaluation method and apparatus for high-resolution remote sensing images, and an electronic device, which can significantly improve the accuracy of quality evaluation of the high-resolution remote sensing images.

In a first aspect, an embodiment of the present invention provides a quality evaluation method for high-resolution remote sensing images, including: obtaining an original high-resolution remote sensing image to be evaluated; generating an apparent reflectivity image based on metadata corresponding to the original high-resolution remote sensing image; carrying out object-oriented image segmentation processing on the original high-resolution remote sensing image based on the apparent reflectivity image to obtain an object-level segmentation image; performing pixel-level quality evaluation on the original high-resolution remote sensing image based on the apparent reflectivity image and a preset decision tree to obtain a pixel-level quality evaluation image; and determining a target quality evaluation image corresponding to the original high-resolution remote sensing image according to the object-level segmentation image and the pixel-level quality evaluation image.

In one embodiment, the metadata includes at least a scaling factor, a distance from the sun to the earth, a solar average irradiance, a solar altitude; generating an apparent reflectivity image based on metadata corresponding to the original high-resolution remote sensing image, wherein the method comprises the following steps: determining a quantization value of a target waveband image from the original high-resolution remote sensing image; determining apparent radiance corresponding to the target waveband image based on the scaling coefficient and the quantization value; and determining an apparent reflectivity image corresponding to the target waveband image based on the apparent radiance, the distance between the day and the earth, the average solar irradiance and the solar altitude angle.

In one embodiment, the target band image comprises at least a green band image, a red band image, and an infrared band image; based on the apparent reflectivity image, carrying out object-oriented image segmentation processing on the original high-resolution remote sensing image to obtain an object-level segmentation image, and the method comprises the following steps: according to the apparent reflectivity images corresponding to the green band image, the red band image and the infrared band image respectively, performing standard false color gray level conversion processing on the original high-resolution remote sensing image to obtain a spectrum image; wherein the number of the wave bands of the spectral image is 1; carrying out nonlinear stretching processing on the spectral image to obtain a normalized spectral image; and carrying out object-oriented image segmentation processing on the normalized spectral image to obtain an object-level segmentation image.

In one embodiment, the metadata further includes a solar zenith angle, a solar azimuth angle and a digital elevation model, the target waveband image includes at least a blue waveband image, a green waveband image, a red waveband image and a near infrared waveband image, and the predetermined decision tree includes one or more of cloud detection, water body detection, cloud shadow detection, mountain shadow detection and vegetation detection; performing pixel-level quality evaluation on the original high-resolution remote sensing image based on the apparent reflectivity image and a preset decision tree to obtain a pixel-level quality evaluation image, which comprises the following steps: performing the cloud detection on the original high-resolution remote sensing image according to the apparent reflectivity images respectively corresponding to the blue band image, the green band image, the red band image and the near-infrared band image to obtain a cloud image; wherein the cloud detection comprises thin cloud detection and/or thick cloud detection; according to the apparent reflectivity images corresponding to the red waveband image and the near infrared waveband image respectively, carrying out the water body detection on the original high-resolution remote sensing image to obtain a water body image; performing the cloud shadow detection on the original high-resolution remote sensing image according to the apparent reflectivity image corresponding to the near-infrared band image to obtain a cloud shadow image; identifying a mountain shadow area in the apparent reflectivity image according to the sun zenith angle, the sun azimuth angle and the digital elevation model to obtain a mountain shadow image; carrying out vegetation detection on the original high-resolution remote sensing image according to the apparent reflectivity images respectively corresponding to the blue waveband image, the red waveband image and the near-infrared waveband image to obtain a vegetation image; determining areas except the cloud image, the water body image, the cloud shadow image, the mountain shadow image and the vegetation image in the original high-resolution remote sensing image as other types of images; and obtaining a pixel-level quality evaluation image based on the cloud image, the water body image, the cloud shadow image, the mountain shadow image, the vegetation image and the other types of images.

In one embodiment, the cloud detection of the original high-resolution remote sensing image according to the apparent reflectance images corresponding to the blue band image, the green band image, the red band image, and the near-infrared band image respectively to obtain a cloud image includes: for each image area in the original high-resolution remote sensing image, calculating an HOT index and a visible light band ratio index according to the apparent reflectivity images respectively corresponding to the blue band image, the green band image and the red band image in the image area; if the HOT index is larger than a HOT segmentation threshold value in the image area, the visible light band ratio index is larger than a ratio segmentation threshold value, and the apparent reflectivity image corresponding to the red band image is larger than a first reflectivity threshold value, determining that the image area is a thick cloud image; extracting the apparent reflectivity image corresponding to the thick cloud image, traversing the thick cloud pixel value corresponding to each wave band in the apparent reflectivity image corresponding to the thick cloud image, and constructing a thick cloud standard spectrum based on the thick cloud pixel value; calculating a spectrum similarity image between the thick cloud image and each non-thick cloud image region in the original high-resolution remote sensing image based on the thick cloud standard spectrum; and for an image area outside the thick cloud image in the original high-resolution remote sensing image, if the spectral similarity image corresponding to the image area is larger than a thin cloud similarity segmentation threshold and the apparent reflectivity image corresponding to the near-infrared band image is larger than a second resolution segmentation threshold, determining that the image area is a thin cloud image.

In one embodiment, the performing the water body detection on the original high-resolution remote sensing image according to the apparent reflectivity images corresponding to the red band image and the near-infrared band image respectively to obtain a water body image includes: for an image area except the cloud image in the original high-resolution remote sensing image, determining a first normalized vegetation index according to the apparent reflectivity images respectively corresponding to the red waveband image and the near-infrared waveband image in the image area; if the first normalized vegetation index in the image area is smaller than a first water body segmentation threshold value and the apparent reflectivity image corresponding to the near-infrared band image is smaller than a third resolution threshold value, determining that the image area is a water body image; or if the first normalized vegetation index is smaller than a second water body segmentation threshold value in the image area and the apparent reflectivity image corresponding to the near-infrared band image is smaller than a fourth resolution threshold value, determining that the image area is a water body image.

In one embodiment, the cloud shadow detection on the original high-resolution remote sensing image according to the apparent reflectivity image corresponding to the near-infrared band image to obtain a cloud shadow image includes: constructing an image value histogram corresponding to the near-infrared band image in an image area except the cloud image and the water body image in the original high-resolution remote sensing image; extracting a target image value of the near-infrared band image corresponding to a first position where the image value is larger than the designated image value in the image value histogram; filling the near-infrared band image based on the target image value; calculating the ratio of the apparent reflectivity image corresponding to the image area to the apparent reflectivity image corresponding to the filled near-infrared band image; and if the ratio is larger than the cloud shadow segmentation threshold, determining that the image area is a cloud shadow image.

In one embodiment, identifying a mountain shadow area in the apparent reflectivity image according to the solar zenith angle, the solar azimuth angle and the digital elevation model, and obtaining a mountain shadow image, includes: for an image area except the cloud image, the water body image and the cloud shadow image in the original high-resolution remote sensing image, determining gradient data and slope data of the position of each pixel in the image area according to the digital elevation model; calculating a mountain shadow probability value corresponding to each pixel according to the sun zenith angle, the sun azimuth angle, the slope data corresponding to each pixel and the slope data; and extracting a target pixel with the mountain shadow probability value larger than a mountain shadow segmentation threshold value from the pixels so as to determine a mountain shadow image based on the target pixel.

In one embodiment, the vegetation detection of the original high-resolution remote sensing image according to the apparent reflectivity images corresponding to the blue band image, the red band image and the near-infrared band image respectively to obtain a vegetation image includes: for image areas except the cloud image, the water body image, the cloud shadow image and the mountain shadow image in the original high-resolution remote sensing image, calculating an enhanced vegetation index according to the apparent reflectivity images respectively corresponding to the blue band image, the red band image and the near-infrared band image in the image areas; calculating a second normalized vegetation index according to the apparent reflectivity images respectively corresponding to the red waveband image and the near-infrared waveband image; calculating a relative activity index according to the apparent reflectivity images respectively corresponding to the red waveband image and the near infrared waveband image; and if the relative vitality index is greater than the vitality index segmentation threshold value, the second normalized vegetation index is greater than the normalized vegetation segmentation threshold value, and the enhanced vegetation index is greater than the enhanced vegetation segmentation threshold value in the image area, determining that the image area is a vegetation image.

In one embodiment, determining a target quality evaluation image corresponding to the original high-resolution remote sensing image according to the object-level segmentation image and the pixel-level quality evaluation image comprises: for each object in the object-level segmentation image, determining the pixel-level quality evaluation image corresponding to the object by using a spatial superposition analysis algorithm; determining the pixel number proportion of each image category in the pixel-level quality evaluation image corresponding to the object; determining an object-level quality evaluation image and a reliability image corresponding to the object according to the image category corresponding to the maximum pixel number proportion; and performing comprehensive evaluation on each object according to the object-level quality evaluation image and the credibility image corresponding to each object by using a weight analysis algorithm to obtain a target quality evaluation image corresponding to the original high-score remote sensing image.

In a second aspect, an embodiment of the present invention further provides a quality evaluation device for high-resolution remote sensing images, including: the image acquisition module is used for acquiring an original high-resolution remote sensing image to be evaluated; the reflectivity generation module is used for generating an apparent reflectivity image based on metadata corresponding to the original high-resolution remote sensing image; the image processing module is used for carrying out object-oriented image segmentation processing on the original high-resolution remote sensing image based on the apparent reflectivity to obtain an object-level segmented image; performing pixel-level quality evaluation on the original high-resolution remote sensing image based on the apparent reflectivity and a preset decision tree to obtain a pixel-level quality evaluation image; and the image evaluation module is used for determining a target quality evaluation image corresponding to the original high-resolution remote sensing image according to the object-level segmentation image and the pixel-level quality evaluation image.

In a third aspect, an embodiment of the present invention further provides an electronic device, which includes a processor and a memory, where the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement any one of the methods provided in the first aspect.

In a fourth aspect, embodiments of the present invention also provide a computer-readable storage medium storing computer-executable instructions that, when invoked and executed by a processor, cause the processor to implement any one of the methods provided in the first aspect.

The quality evaluation method, the quality evaluation device and the electronic equipment for the high-resolution remote sensing image are characterized by firstly obtaining an original high-resolution remote sensing image to be evaluated, generating an apparent reflectivity image based on metadata corresponding to the original high-resolution remote sensing image, performing object-oriented image segmentation processing on the original high-resolution remote sensing image based on the apparent reflectivity image to obtain an object-level segmented image, performing pixel-level quality evaluation on the original high-resolution remote sensing image based on the apparent reflectivity image and a preset decision tree to obtain a pixel-level quality evaluation image, and finally determining a target quality evaluation image corresponding to the original high-resolution remote sensing image according to the object-level segmented image and the pixel-level quality evaluation image. According to the method, firstly, the apparent reflectivity image is generated based on the metadata corresponding to the original high-resolution remote sensing, so that the object-oriented image segmentation processing is respectively carried out on the original high-resolution remote sensing image on the basis of the apparent reflectivity image, and the pixel-level quality evaluation result is carried out by combining the preset decision tree, so that the target quality evaluation image is obtained according to the object-level segmented image and the pixel-level quality evaluation image, and the accuracy of quality evaluation on the high-resolution remote sensing image can be remarkably improved.

Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by the practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.

Fig. 1 is a schematic flow chart of a quality evaluation method for high-resolution remote sensing images according to an embodiment of the present invention;

fig. 2 is a schematic flow chart of another method for evaluating the quality of high-resolution remote sensing images according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of a process for generating an apparent reflectance image according to an embodiment of the present invention;

fig. 4 is a schematic flowchart illustrating an embodiment of generating an object-level segmentation image according to the present invention;

fig. 5 is a schematic flow chart of generating a pixel-level quality evaluation image according to an embodiment of the present invention;

FIG. 6 is a flow chart of segmentation and fusion of a pixel level and an object level according to an embodiment of the present invention;

FIG. 7 is a schematic diagram of an apparent reflectance image according to an embodiment of the present invention;

fig. 8 is a schematic diagram of an object-level segmentation image according to an embodiment of the present invention;

fig. 9 is a schematic diagram of a pixel-level quality evaluation image according to an embodiment of the present invention;

fig. 10 is a schematic diagram of an object-level quality assessment image according to an embodiment of the present invention;

fig. 11 is a schematic diagram of an object-level reliability image according to an embodiment of the present invention;

FIG. 12 is a schematic diagram of a target quality assessment image according to an embodiment of the present invention;

fig. 13 is a schematic structural diagram of a quality evaluation apparatus for high-resolution remote sensing images according to an embodiment of the present invention;

fig. 14 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The high-resolution special project is one of 16 important scientific and technical special projects of medium-long-term scientific and technical development planning compendium (2000-2020), wherein high-resolution one, high-resolution two and high-resolution six (hereinafter abbreviated as GF1, GF2, GF 6) are the most widely applied sub-meter optical satellites, however, the defect of high-resolution image pixel quality evaluation seriously restricts the application of high-resolution satellite images in various fields such as disaster prevention and relief, homeland monitoring, ecological environment protection and the like, and the rapid and accurate image pixel quality evaluation of the high-resolution images becomes the first premise of the application of the high-resolution satellites.

At present, an SC algorithm and an Fmak algorithm both depend on abundant spectral information of Landsat and sentinel2, namely the spectral information can well complete quality evaluation, a GF image only has four bands, the spectral information is very limited, and the evaluation precision must be improved by other technical means; although the object-oriented technology is low in speed, complex in calculation and high in application difficulty, the analysis and interpretation capability of the remote sensing image can be effectively improved, and therefore the quality evaluation accuracy of the remote sensing image can be remarkably improved by using the object-oriented technology and the like.

Based on the above, the invention provides the quality evaluation method and device for the high-resolution remote sensing image and the electronic device, which can obviously improve the accuracy of quality evaluation on the high-resolution remote sensing image.

To facilitate understanding of the present embodiment, first, a method for evaluating quality of a high-resolution remote sensing image disclosed in the embodiment of the present invention is described in detail, referring to a schematic flow chart of a method for evaluating quality of a high-resolution remote sensing image shown in fig. 1, where the method mainly includes the following steps S102 to S108:

and S102, acquiring an original high-resolution remote sensing image to be evaluated. The original high-resolution remote sensing image can also be called original GF multispectral image data, a high-resolution satellite image and the like.

And step S104, generating an apparent reflectivity image based on the metadata corresponding to the original high-resolution remote sensing image. The metadata may include one or more of a scaling factor, a distance between the sun and the earth, a solar average irradiance, a solar altitude, a solar zenith angle, a solar azimuth angle, and a digital elevation model, and may further include data such as a spectral response function, satellite geometric information, sensor information, and image acquisition time. In one embodiment, the scaling coefficient in the metadata of the raw data may be used to generate an apparent radiance image, and then the apparent radiance image is used to calculate and generate an apparent reflectivity image in combination with metadata such as image acquisition time, solar average irradiance, solar elevation angle, and the like.

S106, carrying out object-oriented image segmentation processing on the original high-resolution remote sensing image based on the apparent reflectivity image to obtain an object-level segmentation image; and performing pixel-level quality evaluation on the original high-resolution remote sensing image based on the apparent reflectivity image and a preset decision tree to obtain a pixel-level quality evaluation image. The preset decision tree comprises one or more of cloud detection, water body detection, cloud shadow detection, mountain shadow detection and vegetation detection. In one embodiment, since object-oriented segmentation is more oriented to natural images, and the number of bands of the original high-resolution remote sensing image is more, in order to fully utilize the remote sensing spectrum, a natural pattern spot color grayscale algorithm can be combined, the synthesized bands of the remote sensing standard false color and the apparent reflectivity image are used for calculating grayscale bands, a spectrum image with the number of bands of 1 is generated, a normalized spectrum image is generated through stretching treatment, and finally the normalized spectrum image is subjected to image segmentation to generate an object-level segmented image. In one embodiment, the pixel-level quality assessment method mainly utilizes a decision tree structure to realize the regional detection of the quality of the original high-resolution remote sensing image step by step in a layered manner, specifically comprises the detection of clouds, water bodies, cloud shadows, mountain shadows, vegetation and other types, and finally generates a pixel-level quality assessment image.

And S108, determining a target quality evaluation image corresponding to the original high-resolution remote sensing image according to the object-level segmentation image and the pixel-level quality evaluation image. In an embodiment, although the pixel-level quality assessment image is divided at a high speed, the pixel-level quality assessment image has the problems of broken spots, islands, salt and pepper noises and the like, and the quality assessment precision is reduced, and the object-oriented image division processing has the advantages of accurate image spot boundary, low noise and the like, so that the pixel-level quality assessment image is integrated and optimized by the object-oriented image division processing to construct an object-level quality assessment image, and finally, the object of each region category in the object-level quality assessment image is comprehensively evaluated based on weight analysis, so that the target quality assessment image can be obtained.

According to the quality evaluation method of the high-resolution remote sensing image, firstly, the apparent reflectivity image is generated based on the metadata corresponding to the original high-resolution remote sensing image, so that the original high-resolution remote sensing image is subjected to object-oriented image segmentation processing on the basis of the apparent reflectivity image, and the pixel-level quality evaluation result is carried out by combining the preset decision tree, so that the target quality evaluation image is obtained according to the object-level segmentation image and the pixel-level quality evaluation image, and the accuracy of quality evaluation of the high-resolution remote sensing image can be remarkably improved.

To facilitate understanding of the foregoing embodiments, the embodiment of the present invention provides a schematic flow chart of another method for evaluating quality of a high-resolution remote sensing image, as shown in fig. 2, where the method mainly includes: the method comprises the steps of obtaining an original high-resolution remote sensing image, processing the image to determine an apparent reflectivity image M1, segmenting an object-oriented image to determine an object-level segmentation image M2, determining a pixel-level quality evaluation image M3 based on pixel-level quality evaluation of a decision tree, determining an object-level quality evaluation image M4 based on object-level integration optimization, and determining a target quality evaluation image M5 based on comprehensive evaluation of weight analysis.

Based on the foregoing fig. 2, to facilitate understanding of the foregoing step S104, an embodiment of the present invention provides an implementation manner for generating an apparent reflectance image based on metadata corresponding to an original high-resolution remote sensing image, and refer to a schematic flow chart of generating an apparent reflectance image shown in fig. 3, which mainly includes: the method comprises the steps of obtaining an original high-resolution remote sensing image (original GF data) and metadata, radiometric calibration and apparent reflectivity calculation to determine an apparent reflectivity image M1. For ease of understanding, reference may be made specifically to steps 1 through 3 below:

step 1, determining a quantization value of a target waveband image from an original high-resolution remote sensing image. The target band image may be a blue band image, a green band image, a red band image, a near-infrared band image, or the like. Illustratively, the quantized value of the image element of the ith waveband image in the original high-resolution remote sensing image can be recorded as DN.

And 2, determining the apparent radiance corresponding to the target waveband image based on the scaling coefficient and the quantization value. The scaling coefficients may include gain coefficients and offset coefficients, among others. In one embodiment, the scaling coefficients in the metadata may be used to generate an apparent radiance image, and the specific calculation formula is as follows:

。

wherein the content of the first and second substances,

apparent radiance for band i

(ii) a gain is the gain coefficient of the wave band i, bias is the offset coefficient of the wave band i, and DN is the quantized value of the image element of the i wave band.

And 3, determining an apparent reflectivity image corresponding to the target waveband image based on the apparent radiance, the distance between the day and the ground, the average solar irradiance and the solar altitude angle. In one embodiment, the apparent reflectivity image may be calculated by using the apparent radiance image and combining metadata such as image acquisition time, average solar irradiance, solar altitude, and the like, and the specific formula is as follows:

。

wherein the content of the first and second substances,

the apparent reflectivity of the i wave band is dimensionless; d is the distance between the day and the earth (astronomical units);

is the i-band apparent radiance in units of

；

Is the average solar irradiance outside the atmospheric layer of the wave band in unit of

；

Is the solar altitude angle in degrees.

On the basis of the foregoing fig. 2, in order to facilitate understanding of the foregoing step S106, an embodiment of the present invention provides an implementation that performs object-oriented image segmentation processing on an original high-resolution remote sensing image based on an apparent reflectance image to obtain an object-level segmented image, and referring to a schematic flow chart of generating the object-level segmented image shown in fig. 4, the implementation includes: obtaining an apparent reflectivity image M1, converting standard false color into gray scale, determining a normalized spectrum image by image stretching, and determining an object level segmentation image M2 by image segmentation. For ease of understanding, reference may be made specifically to steps one through three as follows:

according to the apparent reflectivity images corresponding to the green band image, the red band image and the infrared band image respectively, standard false color gray level conversion processing is carried out on the original high-resolution remote sensing image to obtain a spectrum image. Wherein, the number of wave bands of the spectrum image is 1. In one embodiment, the principle of standard false color-to-grayscale processing is as follows:

。

wherein the content of the first and second substances,

the apparent reflectance image corresponding to the green band image,

the apparent reflectance image corresponding to the red band image,

is an apparent reflectivity image corresponding to the infrared band image,

the spectrum image has 1 wave band and the size is (H, W).

And step two, performing nonlinear stretching processing on the spectral image to obtain a normalized spectral image. In one embodiment, the spectral image is processed

Performing 98% nonlinear stretching treatment to generate a normalized spectral image of the fluid 8 data type

Normalized spectral image band

Is 1 and has a size (H, W), the principle of the nonlinear stretching process is as follows:

。

wherein, the first and the second end of the pipe are connected with each other,

is a spectral image

The value of the percentile 2 is,

is a spectral image

The value of percentile 98.

And thirdly, carrying out object-oriented image segmentation processing on the normalized spectral image to obtain an object-level segmented image. In one embodiment, the normalized spectral image may be segmented using the MeanShift image segmentation method

And (4) performing segmentation, wherein MeanShift is an iterative algorithm of non-parameter kernel density estimation, and the core of the iterative algorithm is that sample points of a feature space are clustered, and the sample points converge to a point with zero density gradient, namely a modal point along the gradient rising direction. There are two key parameters for MeanShift, kernel function and window width. Suppose that

I =1, \8230 \8230n, n is a set of any n points in d-dimension Euclidean space and is a kernel functionThe estimated density function for k (x) and window width h is:

。

based on this, the MeanShift transform is then:

。

wherein the content of the first and second substances,

and K (x) is a contour function of K (x) of the kernel function. Normalization of spectral images using MeanShift segmentation model

The segmentation finally obtains an object-level segmentation image M2, where the size of the object-level segmentation image M2 is (H, W), and the number of bands is 1.

On the basis of the foregoing fig. 2, in order to facilitate understanding of the foregoing step S106, an embodiment of the present invention further provides an implementation that performs pixel-level quality assessment on an original high-resolution remote sensing image based on an apparent reflectance image and a preset decision tree to obtain a pixel-level quality assessment image, and refer to a schematic flow diagram for generating a pixel-level quality assessment image shown in fig. 5, where the implementation includes: acquiring an apparent reflectivity image M1; starting cloud detection: judging whether the cloud exists, if so, detecting the thin cloud and the thick cloud based on spectral similarity analysis, determining the thin cloud image and the thick cloud image, and if not, detecting the water body; starting water body detection: judging whether the water body is the water body or not, if so, obtaining a water body image, and if not, carrying out cloud shadow detection; start cloud shadow detection: judging whether the shadow is a cloud shadow, if so, obtaining a cloud shadow image, and if not, performing mountain shadow detection; starting mountain shadow detection: judging whether the shadow is a mountain shadow, if so, obtaining a mountain shadow image, and if not, carrying out vegetation detection; beginning vegetation detection: and judging whether the vegetation is the vegetation or not, if so, obtaining a vegetation image, and if not, obtaining a pixel-level quality evaluation image M3. For ease of understanding, reference may be made specifically to steps a through g as follows:

step a, performing cloud detection on an original high-resolution remote sensing image according to apparent reflectivity images corresponding to a blue band image, a green band image, a red band image and a near-infrared band image respectively to obtain a cloud image; wherein the cloud detection comprises thin cloud detection and/or thick cloud detection. In practical application, a general quality evaluation algorithm can only detect a large class of clouds, and is difficult to distinguish thin clouds from thick clouds, but practice shows that an image part of a thin cloud area can be used for remote sensing application, and cloud detection comprises thick cloud detection and thin cloud detection in order to improve accuracy and fineness of the quality evaluation algorithm. In a specific implementation, see the following steps a1 to a5:

step a1, calculating a HOT index and a visible light band ratio index according to the apparent reflectivity images respectively corresponding to a blue band image, a green band image and a red band image in each image area in an original high-resolution remote sensing image. In one embodiment, thick cloud detection is first performed, and the thick cloud detection is obtained based on threshold segmentation such as HOT index, visible light band ratio (VBR) index, red band, and the like. Wherein, the HOT index calculation formula is as follows:

. Wherein, the first and the second end of the pipe are connected with each other,

the apparent reflectivity image corresponding to the blue band image,

the apparent reflectance image corresponding to the red band image.

The VBR index calculation formula is as follows:

the apparent reflectance image corresponding to the green band image.

Step a2, if the HOT index in the image area is greater than the HOT segmentation threshold, the visible light band ratio index is greater than the ratio segmentation threshold, and the apparent reflectivity image corresponding to the red band image is greater than the first reflectivity threshold, determining that the image area is a thick cloud image. In one embodiment, the thick cloud detection method is as follows:

in order to split the threshold value for the HOT,

the threshold value is divided for the ratio value,

is the first threshold value of the reflectivity and,

is a thick cloud image.

Step a3, extracting an apparent reflectivity image corresponding to the thick cloud image, traversing thick cloud pixel values corresponding to each wave band in the apparent reflectivity image corresponding to the thick cloud image, and constructing a thick cloud standard spectrum based on the thick cloud pixel values. In practical application, through a large number of analysis tests, the threshold value of the thin cloud is lower in the near infrared band relative to the thick cloud, and the thin cloud has certain similarity with the spectrum of the thick cloud. Therefore, firstly, according to the thick cloud area, a thick cloud standard spectrum is constructed

The construction method comprises the following steps:

based on the combination of the apparent reflectivity image M1 and the thick cloud area, the apparent reflectivity image of the thick cloud area is extracted and recorded as

，

Number of wave bands N, go through

Extracting the pixel value of thick cloud from each wave band to obtain data

The dimension size is (N,

），

the number of pixels in the thick cloud region is based on

Construction of a Thick cloud Standard Spectrum

The formula is as follows:

；

；

；

wherein the content of the first and second substances,

is the i wave band

A first one-quartile threshold value is set,

is the i wave band

A threshold value of the number of the three quartiles,

to intercept

Is greater than

And is less than

The data of (a) to (b) to (c),

is the standard spectrum value of the i wave band.

And a4, calculating a spectrum similarity image between the thick cloud image and each non-thick cloud image region in the original high-resolution remote sensing image based on the thick cloud standard spectrum. In one embodiment, the spectral similarity image of the non-thick cloud region and the thick cloud region can be calculated according to the following formula

：

。

Wherein the content of the first and second substances,

is the pixel spectral vector with (i, j) as the position of the non-thick cloud region in the M1 image,

is a standard spectral vector of a thick cloud,

n is the number of wave bands,

is composed of

To the value of the ith band spectrum,

is composed of

And (5) measuring the ith waveband spectral value.

And a5, for an image area except for the thick cloud image in the original high-resolution remote sensing image, if the spectral similarity image corresponding to the image area is larger than the thin cloud similarity segmentation threshold and the apparent reflectivity image corresponding to the near-infrared band image is larger than the second resolution segmentation threshold, determining that the image area is a thin cloud image. In one embodiment, the thin cloud detection method is as follows:

。

wherein the content of the first and second substances,

a threshold value is split for thin cloud similarity,

is an apparent reflectivity image corresponding to the near infrared band image,

a segmentation threshold (i.e., a second resolution segmentation threshold) for the thin cloud in the near infrared band.

Through cloud detection, a cloud image containing thick clouds and thin clouds can be generated

。

And b, performing water body detection on the original high-resolution remote sensing image according to the apparent reflectivity images corresponding to the red wave band image and the near-infrared wave band image respectively to obtain a water body image. In a specific implementation, see the following steps b1 to b2:

step b1, for an image area except for a cloud image in the original high-resolution remote sensing image, determining a first Normalized Difference Vegetation Index (NDVI) according to the apparent reflectivity images respectively corresponding to the red band image and the near-infrared band image in the image area. In one embodiment, the first normalized vegetation index NDVI may be calculated as follows:

。

wherein the content of the first and second substances,

the apparent reflectance image corresponding to the red band image,

the image is the apparent reflectivity image corresponding to the near infrared band image.

B2, if the first normalized vegetation index in the image area is smaller than a first water body segmentation threshold and the apparent reflectivity image corresponding to the near-infrared band image is smaller than a third resolution threshold, determining that the image area is a water body image; or if the first normalized vegetation index in the image area is smaller than the second water body segmentation threshold and the apparent reflectivity image corresponding to the near-infrared band image is smaller than the fourth resolution threshold, determining that the image area is a water body image. In one embodiment, the spectral characteristics of the water body can be used for water body extraction to generate water body images

The method comprises the following steps:

。

wherein the content of the first and second substances,

，

the threshold value of the segmentation of the water body in the NDVI image is, specifically,

a threshold value is segmented for the first body of water,

segmenting a threshold for a second body of water;

，

is a threshold value for dividing the water body in the near infrared band, specifically,

is the third resolution threshold value for the first resolution,

is the fourth resolution threshold.

And c, carrying out cloud shadow detection on the original high-resolution remote sensing image according to the apparent reflectivity image corresponding to the near-infrared band image to obtain a cloud shadow image. In practical application, because cloud shadow is most sensitive in a near-infrared band, cloud shadow detection is carried out based on the apparent reflectivity of the near-infrared band, the near-infrared band is compressed to generate a Uint8 type image, a cumulative histogram (also called as an image value histogram) is constructed for the near-infrared band of a Uint8 data type, and the cumulative histogram is extracted from the near-infrared band of the Uint8 data type

(i.e., designating an image value) and then using the near-infrared image value

Filling the near-infrared band image based on a flood filling algorithm (flood-fill), and finally constructing a ratio index of filling and non-band to detect a cloud shadow area. In a specific implementation, see steps c1 to c5 below:

step c1, constructing an image value histogram corresponding to a near-infrared band image in an image area except a cloud image and a water body image in the original high-resolution remote sensing image. In a specific implementation, the image value histogram may be constructed according to the following formula:

。

wherein the image area is non-cloud and non-waterThe apparent reflectivity image corresponding to the near-infrared band image

Is marked as

。

Step c2, extracting a target image value of the near-infrared band image corresponding to the first position where the specified image value is located in the image value histogram, specifically referring to the following formula:

。

wherein the content of the first and second substances,

for the accumulated histogram data being greater than the specified image value

The near infrared value corresponding to the first position, namely the target image value.

And c3, filling the near-infrared band image based on the target image value, wherein the following formula can be specifically seen:

。

wherein flodfil is a flood fill function,

the filled near infrared band image is the corresponding apparent reflectivity image.

And c4, calculating the ratio of the apparent reflectivity image corresponding to the image area to the apparent reflectivity image corresponding to the filled near-infrared band image.

Step c5, if the ratio is greater than the cloud shadow segmentation threshold, determining that the image area is a cloud shadow image, which can be specifically referred to as the following formula:

。

wherein the content of the first and second substances,

a threshold value is split for the cloud shadow,

is a cloud shadow image.

And d, identifying the mountain shadow area in the apparent reflectivity image according to the solar zenith angle, the solar azimuth angle and the digital elevation model to obtain the mountain shadow image. In practical applications, due to the low GF satellite orbit, for example, when monitoring hilly lands, a mountain shadow area is easily formed on complex terrain with large fluctuation, and in order to improve the accuracy and effectiveness of quality evaluation, it is necessary to detect the mountain shadow of the original high-resolution remote sensing image in the mountain area. In one embodiment, the mountain shadow region in the apparent reflectance image M1 may be identified from the solar zenith angle, the solar azimuth angle, and a Digital Elevation Model DEM (Digital Elevation Model) image in the metadata. Firstly, the gradient of the position of the pixel is calculated by utilizing DEM data (

) And a slope direction (

) Then combining with the sun zenith angle (A), (B)

) And azimuth angle (

) And calculating the mountain shadow probability value of each pixel, and extracting a mountain shadow area by using the mountain shadow probability image constructed by all the pixels and the mountain image threshold value to obtain the influence of the mountain shadow. In a specific implementation, see steps d1 to d3 below:

step d1, determining gradient data and slope data of the position of each pixel in the image area according to the digital elevation model for the image area except the cloud image, the water body image and the cloud shadow image in the original high-resolution remote sensing image. In one embodiment, the slope data and the slope direction data of the position of each pixel can be calculated according to the following formulas:

；

；

；

；

wherein the content of the first and second substances,

and

respectively the difference value of the distance of the DEM in the left-right direction is 1 and the difference value of the distance of the DEM in the up-down direction is 1,

as gradient information

Is the information of the slope direction.

And d2, calculating the mountain shadow probability value corresponding to each pixel according to the sun zenith angle, the sun azimuth angle, and the slope data and the slope direction data corresponding to each pixel. In one embodiment, the mountain shadow probability value corresponding to each pixel element can be calculated according to the following formula:

。

the probability value of the mountain shadow is shown,

is the zenith angle of the sun,

is the solar azimuth.

Step d3, extracting a target pixel with the mountain shadow probability value larger than the mountain shadow segmentation threshold value from the pixels, and determining a mountain shadow image based on the target pixel, wherein the target pixel is specifically shown in the following formula:

。

wherein the content of the first and second substances,

a threshold value is divided for the mountain shadow,

is a mountain shadow image.

And e, carrying out vegetation detection on the original high-resolution remote sensing image according to the apparent reflectivity images corresponding to the blue band image, the red band image and the near-infrared band image respectively to obtain a vegetation image. In practical application, the vegetation region can be extracted according to the spectral characteristics of the vegetation to generate vegetation images

. For ease of understanding, reference may be made specifically to steps e1 to e4 below:

step e1, calculating an Enhanced Vegetation Index (EVI) according to the apparent reflectivity images respectively corresponding to the blue-band image, the red-band image and the near-infrared band image in an image area except the cloud image, the water body image, the cloud shadow image and the mountain shadow image in the original high-resolution remote sensing image. In one embodiment, the enhanced vegetation index EVI may be calculated as follows:

。

the apparent reflectivity image corresponding to the blue band image,

the apparent reflectance image corresponding to the red band image,

And e2, calculating a second normalized vegetation index according to the apparent reflectivity images respectively corresponding to the red wave band image and the near-infrared wave band image. In one embodiment, the second normalized vegetation index NDVI may be calculated as follows:

。

wherein the content of the first and second substances,

the apparent reflectance image corresponding to the red band image,

And e3, calculating the relative vitality index according to the apparent reflectivity images respectively corresponding to the red wave band image and the near infrared wave band image. In one embodiment, the Relative viability Index RVI (Relative Vigor Index) may be calculated according to the following formula:

。

wherein the content of the first and second substances,

the apparent reflectance image corresponding to the red band image,

Step e4, if the relative activity index is greater than the activity index segmentation threshold value, the second normalized vegetation index is greater than the normalized vegetation segmentation threshold value, and the enhanced vegetation index is greater than the enhanced vegetation segmentation threshold value in the image area, determining that the image area is a vegetation image, which can be specifically referred to the following formula:

。

is the cut threshold for vegetation at the RVI index (i.e., vigor index cut threshold),

is the segmentation threshold of the vegetation at the NDVI index (i.e., the normalized vegetation segmentation threshold),

which is the division threshold of vegetation at the EVI index (i.e., the enhanced vegetation division threshold),

is the vegetation image.

And f, determining the regions except the cloud image, the water body image, the cloud shadow image, the mountain shadow image and the vegetation image in the original high-resolution remote sensing image as other types of images. The areas excluding the clouds, water, cloud shadows, mountain shadows and vegetation are other types of images, and are recorded as

。

Step g, based on cloud images, water body images, cloud shadow images, mountain shadow images, vegetation images and other types of imagesAnd obtaining a pixel-level quality evaluation image. In one embodiment, based on

，

，

，

，

And

generating a pixel-level quality evaluation image M3 with the size of (H, W), wherein the pixel value object types in the M3 are shown in the following table 1:

TABLE 1

On the basis of fig. 2, in order to facilitate understanding of step S108, an embodiment of the present invention further provides an implementation method for determining a target quality evaluation image corresponding to an original high-resolution remote sensing image according to an object-level segmented image and a pixel-level quality evaluation image, where the object-level segmented image and the pixel-level quality evaluation image are fused to obtain an object-level quality evaluation image M4, and then the fused object-level quality evaluation image M4 is subjected to comprehensive quality evaluation to obtain a target quality evaluation image M5.

In one embodiment, when fusing the object-level segmented image and the pixel-level quality estimation image to obtain the object-level quality estimation image M4, see a flow chart of pixel-level and object-level segmented fusion as shown in fig. 6, including: the method comprises the steps of obtaining an object level segmentation image M2, all objects of a traversed object level segmentation image M2, obtaining a pixel level quality evaluation image M3, performing spatial superposition analysis, calculating the pixel number proportion and the maximum proportion type of each type in an object C, creating the object type and the proportion of the C, traversing all objects in sequence, and determining an object level quality evaluation image M4 and an object level credibility image P4.

In particular, see the following (1) to (4):

(1) And for each object in the object-level segmentation image, determining a pixel-level quality evaluation image corresponding to the object by using a spatial superposition analysis algorithm. In an embodiment, all objects in the object-level segmentation image M2 are obtained first, and each object is traversed, where one object is C1, and a pixel-level quality assessment image corresponding to an object C1 region is obtained by using spatial superposition analysis in combination with the pixel-level quality assessment image M3, and is denoted as C2.

(2) And determining the proportion of the number of pixels of each image category in the pixel-level quality evaluation image corresponding to the object. In one embodiment, the ratio of the number of pixels in each category in C2 may be calculated, where the ratio of the number of pixels is P at most, and the corresponding category is L, then the attribute of the C1 object category is updated to L, and C1 is L with a confidence level of P, and 0-P < =1.

(3) And determining the object-level quality evaluation image and the credibility image corresponding to the object according to the image category corresponding to the maximum pixel number ratio. In one embodiment, the loop is performed in the above manner to construct the object-level quality assessment image M4 and the confidence image P4, both of which are (H, W).

(4) And comprehensively evaluating each object according to the object-level quality evaluation image and the credibility image corresponding to each object by using a weight analysis algorithm to obtain a target quality evaluation image corresponding to the original high-resolution remote sensing image. In one embodiment, the object-level quality evaluation image M4 and the reliability image P4 are used to perform comprehensive evaluation on each class object in the object-level quality evaluation image M4 based on a weight analysis technique, and finally generate a target quality evaluation image M5. Evaluation score of one of the subjects in the subject-level quality evaluation image M4

The formula is as follows:

。

wherein i is the ith object,

for the basis weights for which the i object is in the L class,

confidence weights for i objects in the L category.

According to the different influence degrees of each category on the quality of the remote sensing image, the basic weight defined by each category is shown in the following table 2:

TABLE 2

According to

And scoring, performing comprehensive evaluation judgment on the quality of each object, and generating a comprehensive quality evaluation value of each object, wherein the specific rule is shown in the following table 3:

table 3:

circulating each object, calculating to obtain the comprehensive evaluation value of each object, and finally generating a quality evaluation image

And finally, removing the speckle noise and the gaps by using a mathematical form to generate a target quality evaluation image M5.

In order to facilitate understanding of the foregoing embodiments, an application example of the quality evaluation method for high-resolution remote sensing images is provided in the embodiments of the present invention. Specifically, the method comprises the following steps:

image processing:

selecting GF2 satellite remote sensing data, wherein the name of one scene image data is GF2_ PMS1_ E109.5_ N29.4_ 20190812/L1A 0004176596-MSS1, the shooting time is 2019, 8, 12 and acquiring spectral response functions, calibration coefficients and satellite geometric angles of each waveband, carrying out radiometric calibration and TOA calculation on the image to obtain an apparent reflectivity image M1, the size is (8784, 7491), and the number of the wavebands is 4. Referring to fig. 7, an apparent reflectance image M1 includes earth features such as thick clouds, thin clouds, cloud shadows, mountain shadows, water bodies, and vegetation, which affect the image quality, and has a certain representativeness.

(II) object-oriented image segmentation:

the apparent reflectance image M1 is subjected to standard false color grayscale conversion processing to obtain a spectral grayscale image with a waveband of 1, the spectral grayscale image is acquired with a minimum value of 0.0006 and a maximum value of 0.4538, and is subjected to normalization processing to generate a normalized image, and an object-level segmentation image M2 is generated based on a MeanShift segmentation algorithm, which is shown in fig. 8 as a schematic diagram of an object-level segmentation image, wherein the size of M2 is (8784, 7491).

And (III) pixel-level quality evaluation based on the decision tree:

the pixel-level quality evaluation method mainly utilizes a decision tree structure to realize area detection influencing GF image quality step by step in a layered manner, specifically comprises detection of clouds, water bodies, cloud shadows, mountain shadows, vegetation and other categories, and the following table 4 is a threshold value of each pixel in an example:

TABLE 4

The pixel-level quality evaluation image M3 is obtained by the above pixel-level threshold segmentation, and referring to a schematic diagram of the pixel-level quality evaluation image shown in fig. 9, the size of the pixel-level quality evaluation image M3 is (8784, 7491).

(IV) segmentation and fusion of pixel level and object level:

based on the object-level segmentation image M2 and the pixel-level quality evaluation image M3, the pixel-level segmentation speed is high, and the object-level segmentation image patch boundary is accurate, because the pixel-level quality evaluation image has the problems of broken patches, isolated islands, salt and pepper noise and the like, and the quality evaluation precision is reduced, the object-oriented segmentation image is used for integrating and optimizing the quality image to obtain an object-level quality evaluation image M4 and an object-level reliability image P4, such as a schematic diagram of an object-level quality evaluation image shown in fig. 10, and a schematic diagram of an object-level reliability image shown in fig. 11.

And (V) comprehensive quality evaluation:

by using the object-level quality evaluation image M4 and the object-level reliability image P4, each class object in the object-level quality evaluation image M4 is comprehensively evaluated based on a weight analysis technique, and a target quality evaluation image is finally generated, which may be specifically shown in a schematic diagram of a target quality evaluation image shown in fig. 12.

In summary, the embodiment of the invention provides a quality evaluation method for high-resolution satellite images, which is fast, accurate and convenient for engineering application, and is suitable for original high-resolution remote sensing images such as GF1, GF2 and GF6, the method utilizes the technologies such as pixel-level and object-level segmentation fusion, decision tree layering, spectral analysis and weight analysis to realize high-resolution image quality evaluation, can detect thin clouds, thick clouds and mountain shadows, and comprehensively evaluate the pixel quality, has the advantages of strong applicability, high accuracy, high automation degree and the like, and can provide basic technical support for the fields such as homeland monitoring, environmental change, disaster monitoring and evaluation and the like.

As to the quality evaluation method of the high-resolution remote sensing image provided by the foregoing embodiment, an embodiment of the present invention provides a quality evaluation device of a high-resolution remote sensing image, referring to a schematic structural diagram of the quality evaluation device of a high-resolution remote sensing image shown in fig. 13, and the device mainly includes the following components:

an image obtaining module 1302, configured to obtain an original high-resolution remote sensing image to be evaluated;

the reflectivity generation module 1304 is used for generating an apparent reflectivity image based on metadata corresponding to the original high-resolution remote sensing image;

the image processing module 1306 is used for performing object-oriented image segmentation processing on the original high-resolution remote sensing image based on the apparent reflectivity to obtain an object-level segmented image; performing pixel-level quality evaluation on the original high-resolution remote sensing image based on the apparent reflectivity and a preset decision tree to obtain a pixel-level quality evaluation image;

the image evaluation module 1308 is configured to determine a target quality evaluation image corresponding to the original high-resolution remote sensing image according to the object-level segmentation image and the pixel-level quality evaluation image.

According to the quality evaluation device of the high-resolution remote sensing image, the apparent reflectivity image is generated based on the metadata corresponding to the original high-resolution remote sensing image, the original high-resolution remote sensing image is subjected to object-oriented image segmentation processing on the basis of the apparent reflectivity image, the pixel-level quality evaluation result is carried out by combining the preset decision tree, and the target quality evaluation image is obtained according to the object-level segmentation image and the pixel-level quality evaluation image, so that the accuracy of quality evaluation of the high-resolution remote sensing image can be remarkably improved.

In one embodiment, the metadata includes at least a scaling factor, a distance from the sun to the earth, a solar average irradiance, a solar altitude; the reflectivity generation module 1304 is also for: determining a quantization value of a target waveband image from the original high-resolution remote sensing image; determining apparent radiance corresponding to the target waveband image based on the scaling coefficient and the quantization value; and determining an apparent reflectivity image corresponding to the target waveband image based on the apparent radiance, the distance between the day and the ground, the average solar irradiance and the solar altitude angle.

In one embodiment, the target band image comprises at least a green band image, a red band image, and an infrared band image; the image processing module 1306 is further configured to: according to the apparent reflectivity images corresponding to the green wave band image, the red wave band image and the infrared wave band image respectively, performing standard false color gray level conversion processing on the original high-resolution remote sensing image to obtain a spectrum image; wherein the number of wave bands of the spectral image is 1; carrying out nonlinear stretching processing on the spectral image to obtain a normalized spectral image; and carrying out object-oriented image segmentation processing on the normalized spectral image to obtain an object-level segmentation image.

In one embodiment, the metadata further includes a solar zenith angle, a solar azimuth angle and a digital elevation model, the target waveband image includes at least a blue waveband image, a green waveband image, a red waveband image and a near-infrared waveband image, and the predetermined decision tree includes one or more of cloud detection, water body detection, cloud shadow detection, mountain shadow detection and vegetation detection; the image processing module 1306 is further configured to: performing cloud detection on the original high-resolution remote sensing image according to the apparent reflectivity images respectively corresponding to the blue band image, the green band image, the red band image and the near-infrared band image to obtain a cloud image; wherein the cloud detection comprises thin cloud detection and/or thick cloud detection; according to the apparent reflectivity images corresponding to the red wave band image and the near infrared wave band image respectively, carrying out water body detection on the original high-resolution remote sensing image to obtain a water body image; carrying out cloud shadow detection on the original high-resolution remote sensing image according to the apparent reflectivity image corresponding to the near-infrared band image to obtain a cloud shadow image; identifying a mountain shadow area in the apparent reflectivity image according to the solar zenith angle, the solar azimuth angle and the digital elevation model to obtain a mountain shadow image; carrying out vegetation detection on the original high-resolution remote sensing image according to the apparent reflectivity images respectively corresponding to the blue band image, the red band image and the near-infrared band image to obtain a vegetation image; determining areas except for cloud images, water body images, cloud shadow images, mountain shadow images and vegetation images in the original high-resolution remote sensing images as other types of images; based on the cloud image, the water body image, the cloud shadow image, the mountain shadow image, the vegetation image and other types of images, the pixel-level quality evaluation image is obtained.

In one embodiment, the image processing module 1306 is further configured to: for each image area in the original high-resolution remote sensing image, calculating a HOT index and a visible light band ratio index according to the apparent reflectivity images respectively corresponding to the blue band image, the green band image and the red band image in the image area; if the HOT index is larger than the HOT segmentation threshold value in the image area, the visible light wave band ratio index is larger than the ratio segmentation threshold value, and the apparent reflectivity image corresponding to the red wave band image is larger than the first reflectivity threshold value, determining that the image area is a thick cloud image; extracting an apparent reflectivity image corresponding to the thick cloud image, traversing a thick cloud pixel value corresponding to each wave band in the apparent reflectivity image corresponding to the thick cloud image, and constructing a thick cloud standard spectrum based on the thick cloud pixel value; calculating a spectrum similarity image between the thick cloud image and each non-thick cloud image region in the original high-resolution remote sensing image based on the thick cloud standard spectrum; and for an image area outside the thick cloud image in the original high-resolution remote sensing image, if the spectral similarity image corresponding to the image area is larger than the thin cloud similarity segmentation threshold and the apparent reflectivity image corresponding to the near-infrared band image is larger than the second resolution segmentation threshold, determining that the image area is a thin cloud image.

In one embodiment, the image processing module 1306 is further configured to: for an image area except for a cloud image in an original high-resolution remote sensing image, determining a first normalized vegetation index according to the apparent reflectivity images respectively corresponding to a red wave band image and a near-infrared wave band image in the image area; if the first normalized vegetation index in the image area is smaller than a first water body segmentation threshold value and the apparent reflectivity image corresponding to the near-infrared band image is smaller than a third resolution threshold value, determining that the image area is a water body image; or if the first normalized vegetation index in the image area is smaller than the second water body segmentation threshold and the apparent reflectivity image corresponding to the near-infrared band image is smaller than the fourth resolution threshold, determining that the image area is a water body image.

In one embodiment, the image processing module 1306 is further configured to: constructing an image value histogram corresponding to a near-infrared band image in an image area except a cloud image and a water body image in an original high-resolution remote sensing image; extracting a target image value of the near-infrared band image corresponding to a first position where the image value is larger than the specified image value in the image value histogram; filling the near-infrared band image based on the target image value; calculating the ratio of the apparent reflectivity image corresponding to the image area to the apparent reflectivity image corresponding to the filled near-infrared band image; and if the ratio is larger than the cloud shadow segmentation threshold, determining the image area as a cloud shadow image.

In one embodiment, the image processing module 1306 is further configured to: for an image area except for a cloud image, a water body image and a cloud shadow image in an original high-resolution remote sensing image, determining gradient data and slope data of the position of each pixel in the image area according to a digital elevation model; calculating the probability value of the mountain shadow corresponding to each pixel according to the sun zenith angle, the sun azimuth angle, the slope data corresponding to each pixel and the slope data; and extracting a target pixel with the mountain shadow probability value larger than the mountain shadow segmentation threshold value from the pixels so as to determine a mountain shadow image based on the target pixel.

In one embodiment, the image processing module 1306 is further configured to: for image areas except for a cloud image, a water body image, a cloud shadow image and a mountain shadow image in the original high-resolution remote sensing image, calculating an enhanced vegetation index according to the apparent reflectivity images respectively corresponding to a blue band image, a red band image and a near-infrared band image in the image areas; calculating a second normalized vegetation index according to the apparent reflectivity images respectively corresponding to the red wave band image and the near-infrared wave band image; calculating a relative activity index according to the apparent reflectivity images respectively corresponding to the red wave band image and the near infrared wave band image; and if the relative vitality index is greater than the vitality index segmentation threshold value in the image area, the second normalized vegetation index is greater than the normalized vegetation segmentation threshold value, and the enhanced vegetation index is greater than the enhanced vegetation segmentation threshold value, determining that the image area is a vegetation image.

In one embodiment, the image evaluation module 1308 is further configured to: for each object in the object-level segmentation image, determining a pixel-level quality evaluation image corresponding to the object by using a spatial superposition analysis algorithm; determining the proportion of the number of pixels of each image category in the pixel-level quality evaluation image corresponding to the object; determining an object-level quality evaluation image and a reliability image corresponding to the object according to the image category corresponding to the maximum pixel number proportion; and performing comprehensive evaluation on each object according to the object-level quality evaluation image and the credibility image corresponding to each object by using a weight analysis algorithm to obtain a target quality evaluation image corresponding to the original high-resolution remote sensing image.

The device provided by the embodiment of the present invention has the same implementation principle and technical effect as the method embodiments, and for the sake of brief description, reference may be made to the corresponding contents in the method embodiments without reference to the device embodiments.

The embodiment of the invention provides electronic equipment, which particularly comprises a processor and a storage device; the storage means has stored thereon a computer program which, when executed by the processor, performs the method of any of the above described embodiments.

Fig. 14 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, where the electronic device 100 includes: processor 140, memory 141, bus 142 and communication interface 143, said processor 140, communication interface 143 and memory 141 being connected by bus 142; processor 140 is operative to execute executable modules, such as computer programs, stored in memory 141.

The memory 141 may include a high-speed Random Access Memory (RAM) and may further include a non-volatile memory (non-volatile memory), such as at least one disk memory. The communication connection between the network element of the system and at least one other network element is implemented through at least one communication interface 143 (which may be wired or wireless), and the internet, a wide area network, a local network, a metropolitan area network, and the like may be used.

The bus 142 may be an ISA bus, PCI bus, EISA bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one double-headed arrow is shown in FIG. 14, but that does not indicate only one bus or one type of bus.

The memory 141 is used for storing a program, the processor 140 executes the program after receiving an execution instruction, and the method performed by the apparatus defined by the flow program disclosed in any of the foregoing embodiments of the present invention may be applied to the processor 140, or implemented by the processor 140.

The processor 140 may be an integrated circuit chip having signal processing capabilities. In implementation, the steps of the above method may be performed by integrated logic circuits of hardware or instructions in the form of software in the processor 140. The Processor 140 may be a general-purpose Processor, and includes a Central Processing Unit (CPU), a Network Processor (NP), and the like; the device can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-programmable gate Array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The various methods, steps, and logic blocks disclosed in the embodiments of the present invention may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in connection with the embodiments of the present invention may be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules may be located in ram, flash, rom, prom, or eprom, registers, etc. as is well known in the art. The storage medium is located in the memory 141, and the processor 140 reads the information in the memory 141, and completes the steps of the above method in combination with the hardware thereof.

The computer program product of the readable storage medium provided in the embodiment of the present invention includes a computer readable storage medium storing a program code, where instructions included in the program code may be used to execute the method described in the foregoing method embodiment, and specific implementation may refer to the foregoing method embodiment, which is not described herein again.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, and various media capable of storing program codes.

Finally, it should be noted that: the above-mentioned embodiments are only specific embodiments of the present invention, which are used for illustrating the technical solutions of the present invention and not for limiting the same, and the protection scope of the present invention is not limited thereto, although the present invention is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope of the present disclosure; such modifications, changes or substitutions do not depart from the spirit and scope of the embodiments of the present invention, and they should be construed as being included therein. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A quality evaluation method of high-resolution remote sensing images is characterized by comprising the following steps:

obtaining an original high-resolution remote sensing image to be evaluated;

generating an apparent reflectivity image based on metadata corresponding to the original high-resolution remote sensing image;

performing object-oriented image segmentation processing on the original high-resolution remote sensing image based on the apparent reflectivity image to obtain an object-level segmented image; performing pixel-level quality evaluation on the original high-resolution remote sensing image based on the apparent reflectivity image and a preset decision tree to obtain a pixel-level quality evaluation image;

and determining a target quality evaluation image corresponding to the original high-resolution remote sensing image according to the object-level segmentation image and the pixel-level quality evaluation image.

2. The method for evaluating the quality of high-resolution remote sensing images according to claim 1, wherein the metadata at least comprises a scaling coefficient, a distance between the day and the earth, an average irradiance of the sun, and a solar altitude; generating an apparent reflectivity image based on metadata corresponding to the original high-resolution remote sensing image, wherein the method comprises the following steps:

determining a quantization value of a target waveband image from the original high-resolution remote sensing image;

determining apparent radiance corresponding to the target waveband image based on the scaling coefficient and the quantization value;

and determining an apparent reflectivity image corresponding to the target waveband image based on the apparent radiance, the distance between the day and the ground, the average solar irradiance and the solar altitude angle.

3. The quality evaluation method of the high-resolution remote sensing image according to claim 2, wherein the target waveband image at least comprises a green waveband image, a red waveband image and an infrared waveband image; carrying out object-oriented image segmentation processing on the original high-resolution remote sensing image based on the apparent reflectivity image to obtain an object-level segmentation image, wherein the object-level segmentation image comprises the following steps:

according to the apparent reflectivity images corresponding to the green band image, the red band image and the infrared band image respectively, performing standard false color gray level conversion processing on the original high-resolution remote sensing image to obtain a spectrum image; wherein the number of wave bands of the spectral image is 1;

carrying out nonlinear stretching processing on the spectral image to obtain a normalized spectral image;

and carrying out object-oriented image segmentation processing on the normalized spectral image to obtain an object-level segmentation image.

4. The quality assessment method for the high-resolution remote sensing image according to claim 2, wherein the metadata further comprises a solar zenith angle, a solar azimuth angle and a digital elevation model, the target waveband image at least comprises a blue waveband image, a green waveband image, a red waveband image and a near-infrared waveband image, and the preset decision tree comprises one or more of cloud detection, water body detection, cloud shadow detection, mountain shadow detection and vegetation detection;

performing pixel-level quality evaluation on the original high-resolution remote sensing image based on the apparent reflectivity image and a preset decision tree to obtain a pixel-level quality evaluation image, which comprises the following steps:

performing the cloud detection on the original high-resolution remote sensing image according to the apparent reflectivity images respectively corresponding to the blue band image, the green band image, the red band image and the near-infrared band image to obtain a cloud image; wherein the cloud detection comprises thin cloud detection and/or thick cloud detection;

according to the apparent reflectivity images respectively corresponding to the red waveband image and the near infrared waveband image, carrying out the water body detection on the original high-resolution remote sensing image to obtain a water body image;

performing the cloud shadow detection on the original high-resolution remote sensing image according to the apparent reflectivity image corresponding to the near-infrared band image to obtain a cloud shadow image;

identifying a mountain shadow area in the apparent reflectivity image according to the solar zenith angle, the solar azimuth angle and the digital elevation model to obtain a mountain shadow image;

carrying out vegetation detection on the original high-resolution remote sensing image according to the apparent reflectivity images respectively corresponding to the blue waveband image, the red waveband image and the near-infrared waveband image to obtain a vegetation image;

determining areas except the cloud image, the water body image, the cloud shadow image, the mountain shadow image and the vegetation image in the original high-resolution remote sensing image as other types of images;

and obtaining a pixel-level quality evaluation image based on the cloud image, the water body image, the cloud shadow image, the mountain shadow image, the vegetation image and the other types of images.

5. The quality evaluation method of the high-resolution remote sensing image according to claim 4, wherein the cloud detection is performed on the original high-resolution remote sensing image according to the apparent reflectance images corresponding to the blue band image, the green band image, the red band image and the near-infrared band image, respectively, to obtain a cloud image, and the method comprises:

for each image area in the original high-resolution remote sensing image, calculating a HOT index and a visible light band ratio index according to the apparent reflectivity images respectively corresponding to the blue band image, the green band image and the red band image in the image area;

if the HOT index is larger than a HOT segmentation threshold value in the image area, the visible light band ratio index is larger than a ratio segmentation threshold value, and the apparent reflectivity image corresponding to the red band image is larger than a first reflectivity threshold value, determining that the image area is a thick cloud image;

extracting the apparent reflectivity image corresponding to the thick cloud image, traversing the thick cloud pixel value corresponding to each wave band in the apparent reflectivity image corresponding to the thick cloud image, and constructing a thick cloud standard spectrum based on the thick cloud pixel value;

calculating a spectrum similarity image between the thick cloud image and each non-thick cloud image region in the original high-resolution remote sensing image based on the thick cloud standard spectrum;

and for an image area outside the thick cloud image in the original high-resolution remote sensing image, if the spectral similarity image corresponding to the image area is larger than a thin cloud similarity segmentation threshold and the apparent reflectivity image corresponding to the near-infrared band image is larger than a second resolution segmentation threshold, determining that the image area is a thin cloud image.

6. The quality evaluation method of the high-resolution remote sensing image according to claim 4, wherein the water body detection is performed on the original high-resolution remote sensing image according to the apparent reflectivity images corresponding to the red band image and the near-infrared band image, respectively, to obtain a water body image, and the method comprises the following steps:

for an image area except the cloud image in the original high-resolution remote sensing image, determining a first normalized vegetation index according to the apparent reflectivity images respectively corresponding to the red waveband image and the near-infrared waveband image in the image area;

if the first normalized vegetation index in the image area is smaller than a first water body segmentation threshold value and the apparent reflectivity image corresponding to the near-infrared band image is smaller than a third resolution threshold value, determining that the image area is a water body image;

or if the first normalized vegetation index is smaller than a second water body segmentation threshold value in the image area and the apparent reflectivity image corresponding to the near-infrared band image is smaller than a fourth resolution threshold value, determining that the image area is a water body image.

7. The quality evaluation method of the high-resolution remote sensing image according to claim 4, wherein the cloud shadow detection is performed on the original high-resolution remote sensing image according to the apparent reflectivity image corresponding to the near-infrared band image to obtain a cloud shadow image, and the method comprises the following steps:

constructing an image value histogram corresponding to the near-infrared band image in an image area except the cloud image and the water body image in the original high-resolution remote sensing image;

extracting a target image value of the near-infrared band image corresponding to a first position where the image value is larger than the designated image value in the image value histogram;

filling the near-infrared band image based on the target image value;

calculating the ratio of the apparent reflectivity image corresponding to the image area to the apparent reflectivity image corresponding to the filled near-infrared band image;

and if the ratio is greater than the cloud shadow segmentation threshold, determining that the image area is a cloud shadow image.

8. The quality evaluation method for the high-resolution remote sensing image according to claim 4, wherein the identifying of the mountain shadow area in the apparent reflectance image according to the solar zenith angle, the solar azimuth angle and the digital elevation model to obtain the mountain shadow image comprises:

for an image area except the cloud image, the water body image and the cloud shadow image in the original high-resolution remote sensing image, determining slope data and slope data of the position of each pixel in the image area according to the digital elevation model;

calculating a mountain shadow probability value corresponding to each pixel according to the sun zenith angle, the sun azimuth angle, the slope data corresponding to each pixel and the slope data;

and extracting a target pixel with the mountain shadow probability value larger than a mountain shadow segmentation threshold value from the pixels so as to determine a mountain shadow image based on the target pixel.

9. The quality assessment method for the high-resolution remote sensing image according to claim 4, wherein the vegetation detection is performed on the original high-resolution remote sensing image according to the apparent reflectivity images corresponding to the blue band image, the red band image and the near-infrared band image respectively to obtain a vegetation image, and the method comprises the following steps:

calculating an enhanced vegetation index according to the apparent reflectivity images respectively corresponding to the blue band image, the red band image and the near infrared band image in an image area except the cloud image, the water body image, the cloud shadow image and the mountain shadow image in the original high-resolution remote sensing image;

calculating a second normalized vegetation index according to the apparent reflectivity images respectively corresponding to the red waveband image and the near infrared waveband image;

calculating a relative activity index according to the apparent reflectivity images respectively corresponding to the red waveband image and the near infrared waveband image;

and if the relative vitality index is greater than the vitality index segmentation threshold value, the second normalized vegetation index is greater than the normalized vegetation segmentation threshold value, and the enhanced vegetation index is greater than the enhanced vegetation segmentation threshold value in the image area, determining that the image area is a vegetation image.

10. The quality evaluation method of the high-resolution remote sensing image according to claim 1, wherein determining a target quality evaluation image corresponding to the original high-resolution remote sensing image according to the object-level segmentation image and the pixel-level quality evaluation image comprises:

for each object in the object-level segmentation image, determining the pixel-level quality evaluation image corresponding to the object by using a spatial superposition analysis algorithm;

determining the proportion of the number of pixels of each image category in the pixel-level quality evaluation image corresponding to the object;

determining an object-level quality evaluation image and a reliability image corresponding to the object according to the image category corresponding to the maximum pixel number proportion;

and comprehensively evaluating each object according to the object-level quality evaluation image and the credibility image corresponding to each object by using a weight analysis algorithm to obtain a target quality evaluation image corresponding to the original high-score remote sensing image.

11. A quality evaluation device for high-resolution remote sensing images is characterized by comprising:

the image acquisition module is used for acquiring an original high-resolution remote sensing image to be evaluated;

the reflectivity generation module is used for generating an apparent reflectivity image based on metadata corresponding to the original high-resolution remote sensing image;

the image processing module is used for carrying out object-oriented image segmentation processing on the original high-resolution remote sensing image based on the apparent reflectivity to obtain an object-level segmentation image; performing pixel-level quality evaluation on the original high-resolution remote sensing image based on the apparent reflectivity and a preset decision tree to obtain a pixel-level quality evaluation image;

and the image evaluation module is used for determining a target quality evaluation image corresponding to the original high-resolution remote sensing image according to the object-level segmentation image and the pixel-level quality evaluation image.

12. An electronic device comprising a processor and a memory, the memory storing computer-executable instructions executable by the processor, the processor executing the computer-executable instructions to implement the method of any one of claims 1 to 10.

13. A computer-readable storage medium having computer-executable instructions stored thereon which, when invoked and executed by a processor, cause the processor to perform the method of any of claims 1 to 10.