WO2024007598A1

WO2024007598A1 - Accurate screening method for remote sensing image group

Info

Publication number: WO2024007598A1
Application number: PCT/CN2023/077837
Authority: WO
Inventors: 贾若愚; 陈宇; 李洁; 段红伟; 邹圣兵
Original assignee: 北京数慧时空信息技术有限公司
Priority date: 2022-07-04
Filing date: 2023-02-23
Publication date: 2024-01-11
Also published as: CN114882379A; CN114882379B

Abstract

The present invention relates to the technical field of remote sensing. Provided is an accurate screening method for a remote sensing image group. The method comprises: selecting a target area, acquiring an initial image group within the target area, and calculating test scores of the initial image group; dividing the initial image group into a first image group and a second image group according to the test scores; expanding the first image group on the basis of a first expansion policy to obtain a third image group; calculating evaluation scores of the third image group, and performing screening on the third image group according to the evaluation scores to obtain an optimal image group; and expanding the optimal image group according to a second expansion policy to obtain a final image group. The present invention can greatly save on manpower, material resources and financial resources, and achieve the aim of quickly and accurately selecting images required by users.

Description

An accurate screening method for remote sensing image groups

Technical field

The invention relates to the field of remote sensing technology, and specifically relates to an accurate screening method for remote sensing image groups.

Background technique

The earth is the common home of human existence. With the continuous progress of human civilization, the use of technological means to continuously discover and understand the unknown world has become a powerful driving force for the progress of human civilization. Due to the vastness and vastness of the earth's surface, although humans have evolved on the earth for tens of millions of years, their understanding of the environment in which they live is very limited, from both local and overall aspects. It was not until the mid-20th century with the emergence of satellite remote sensing technology that humans obtained image data of the earth's surface through the "eye in the sky" that the relatively continuous understanding of the earth as a whole truly opened. Especially in the early 21st century, with the rapid advancement of technology and the vigorous development of remote sensing technology, the number of remote sensing satellites continues to increase, and the observation data of the earth's surface obtained by humans continue to accumulate rapidly. The data scale has now reached the EB level, and subsequently applied The emerging big data technology provides technical support for the processing of massive data and information mining.

Space Earth observation technology provides multi-temporal, wide-coverage, three-dimensional remote sensing image data for Earth system scientific research, making it possible to observe, understand, simulate and predict the behavior of the entire Earth system. Remote sensing image data obtained through satellites, aviation and other means have rich spatial, temporal and attribute information, and have become an important source for studying and solving key issues such as global change, disaster prevention and reduction, and sustainable development. Traditional remote sensing data screening methods are mostly based on query engines provided by spatial databases to search for qualified results from massive data. However, this traditional remote sensing data screening method has the following problems. On the one hand, the setting of query conditions is relatively fixed. On the other hand, there is a lot of redundancy in the filtered remote sensing data, and manual selection is needed to obtain the optimal remote sensing data. Data is very time-consuming, labor-intensive and inefficient. Therefore, how to quickly and automatically screen out the optimal remote sensing image data and reduce the workload and time of manual selection is particularly important.

Contents of the invention

Based on the above technical problems, the present invention provides a method for obtaining a set of images of moderate quantity and high quality for the target area selected by the user through the steps of preliminary screening, expansion, redundancy removal and expansion. The search and screening process of this method It is very fast, can meet the needs of users, and greatly saves manpower, material and financial resources.

The present invention provides a method for accurate screening of remote sensing image groups. The method includes:

S1 selects the target area, obtains the initial image group within the target area, and calculates the detection score of the initial image group;

S2 divides the initial image group into a first image group and a second image group according to the detection score;

S3 expands the first image group based on the first expansion strategy to obtain a third image group;

S4 Calculate the evaluation score of the third image group, screen the third image group according to the evaluation score, and obtain the preferred image group;

S5 expands the preferred image group according to the second expansion strategy to obtain the final image group.

Optionally, step S1 includes:

Perform cloud cover detection on the images in the initial image group to obtain cloud content;

Perform quality item detection on the images in the initial image group to obtain a quality score, where the quality item detection includes strip detection, high exposure detection, edge detection and histogram detection;

The cloud content and the quality score are integrated to obtain a detection score, where the value range of the detection score is 0-100.

Optionally, step S2 includes:

Sort the initial image group from high to low according to the detection score, start filtering from the positive order, until the coverage of the union of the filtered images on the target area reaches the coverage threshold, and use the filtered image as the first image group, the images after excluding the first image group from the initial image group are used as the second image group.

Optionally, the quality score is obtained by integrating and calculating the detection results of each quality item, and the value range of the quality score is 0-100.

Optionally, step S3 includes:

Use one image in the first image group as the image to be expanded;

Determine the area to be expanded in the image to be expanded;

The second image group is sorted from high to low according to the detection score, and the images in the second image group are selected from the positive order to expand the area to be expanded until the selected second image group is The union of images reaches 100% coverage of the area to be expanded;

Traverse all the images in the first image group, and combine the selected images in the second image group with the first image group as a third image group.

Optionally, the image to be expanded intersects with a union of the remaining images in the first image group except the image to be expanded, and the image to be expanded is divided into an intersecting area and a non-intersecting area, and the non-intersecting area is the area to be expanded.

Optionally, step S4 includes:

Obtain the metadata of the initial image group, and parse the metadata to obtain time series data and sensor data;

Calculate the variance of the time series data and the sensor data, and integrate the quality score, the variance of the time series data, and the variance of the sensor data into a comprehensive score;

Construct an evaluation function based on the cloud content and the comprehensive score;

Using the evaluation function to calculate the third image group, obtain an evaluation score of the third image group;

The third image group is sorted from low to high according to its evaluation score, and screened from positive order according to the screening strategy to obtain the preferred image group.

Optionally, the screening strategy is:

Step 1: Select a single image in sequence starting from the main sequence as a pre-screened image, and use the image set in the third image group excluding the pre-screened image as the remaining image group;

Step 2: Set a retention rule. The retention rule is that if the coverage of the target area by the union of the images in the remaining image group decreases after removing the pre-screened image, then the corresponding pre-screened image will be removed. The pre-screened images are retained. If the pre-screened images do not meet the retention rules, the third step is executed;

Step 3: Calculate the average evaluation score of the third image group and the average evaluation score of the remaining image group. If the average evaluation score of the remaining image group is lower than the average evaluation score of the third image group, then Keep the corresponding pre-screened images;

Step 4: If two or more images among the images retained after performing steps 2 and 3 have the same evaluation scores, the images with the same evaluation scores will be ranked in descending order of cloud content. Sort to highest, retain the first image, and filter out the remaining images.

Optionally, step S5 includes:

Use images in the initial image group other than the preferred image group as candidate image groups;

The candidate image group is determined based on the second expansion strategy, and images that meet the requirements of the second expansion strategy are classified into the expanded image group;

The preferred image group is expanded using the expanded image group to obtain a final image group.

Optionally, the candidate image group is determined based on the second expansion strategy, and the images that meet the requirements of the second expansion strategy are classified into the expanded image group, including:

Compare one image in the candidate image group with all the images in the preferred image group one by one,

If there is no image in the preferred image group that covers 100% of the image in the candidate image group, then calculate the evaluation score of the image in the candidate image group. If it is higher than the preferred image The average evaluation score of all images in the group, then the image in the candidate image group is classified into the extended image group;

If there is an image in the preferred image group that covers 100% of the image in the candidate image group, then the evaluation scores of the two images are calculated. If the evaluation score of the image in the candidate image group is If the score is higher than the evaluation score of the image in the preferred image group, then the image in the candidate image group is classified into the extended image group;

If there are two or more image groups in the preferred image group with a coverage rate of 100% for the image in the candidate image group, then compare the evaluation score of the image in the candidate image group with The evaluation score of each image in the image group. If the evaluation score of the image in the candidate image group is higher than the lowest evaluation score in the image group, then the image in the candidate image group into the extended image group.

The beneficial effects of the present invention are: the present invention provides an accurate screening method for remote sensing image groups. The method includes: selecting a target area, obtaining an initial image group within the target area, and calculating the detection score of the initial image group; The detection score divides the initial image group into a first image group and a second image group; the first image group is expanded based on the first expansion strategy to obtain a third image group; and the evaluation of the third image group is calculated The third image group is screened according to the evaluation score to obtain the preferred image group; the preferred image group is expanded according to the second expansion strategy to obtain the final image group. The images were overall evaluated and screened based on cloud content, comprehensive scores, and evaluation scores. After initial screening, first expansion, fine screening, and second expansion, the final image group was obtained with the optimal number and better image quality. High, it can greatly improve the accuracy of images when used for interpretation of target areas and other applications. Moreover, the screening process of the present invention is fast and the amount of calculation is not large. It can greatly save manpower, material and financial resources, and achieve rapid and accurate screening required by users. The purpose of the image.

Description of the drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the drawings needed to be used in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some of the drawings of the present invention. Embodiments, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without exerting creative efforts.

Figure 1 is a method flow chart according to an embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all the embodiments. It should be noted that as long as there is no conflict, the various embodiments of the present invention and the various features of the embodiments can be combined with each other, and the resulting technical solutions are within the protection scope of the present invention.

It should be noted that the terms used herein are for the purpose of describing specific embodiments only, and are not intended to limit the exemplary embodiments according to the present invention. As used herein, the singular forms are also intended to include the plural forms unless the context clearly indicates otherwise. Furthermore, it will be understood that when the terms "comprises" and/or "includes" are used in this specification, they indicate There are features, steps, operations, means, components and/or combinations thereof.

Please refer to Figure 1. The present invention provides a precise screening method for remote sensing image groups. The method includes:

According to the images of the required area provided by the user, the target area is determined, and all images in the target area are obtained to form an initial image group. The images in the initial image group are original image data.

Perform preliminary detection on all images in the initial image group to obtain detection scores. Specifically:

S11 Perform cloud cover detection on the images in the initial image group. The cloud cover detection method in this embodiment can be:

Step 1: Determine whether the image in the initial image group is a panchromatic image or a multispectral image. If it is a panchromatic image, go directly to step 2. If it is a multispectral image, convert the multispectral image in the initial image group into a single band. Luminance image,conversion is achieved by the following formula:

In the formula, P ₍ _i,j ₎ represents the brightness value of the pixel located at (i,j) in the converted brightness map, R ₍ _i,j ₎ , G ₍ _i,j ₎ , B ₍ _i,j ₎ respectively Represents the brightness value of the red, green and blue bands of the pixel located at ( i,j ) in the multispectral image. Since clouds exhibit Mie scattering of sunlight, they have relatively strong scattering in each band. As shown in the multispectral image, the clouds are white and the brightness values in each band are very high. Non-cloud surface targets diffusely reflect sunlight, and the reflectivity in different bands is often different. When presented on multispectral images, non-cloud targets are often colored and have different brightness values in each band, but the minimum brightness value is often low. Therefore, the single-band brightness image obtained by extracting the minimum value of each band brightness value through Equation (1) combines the brightness and saturation information of the multispectral image, making it easier to distinguish clouds from non-cloud targets.

Step 2: Rough estimate of dual brightness thresholds: Calculate the corresponding highest and lowest brightness thresholds based on cloud-free and cloud-containing images. Calculating the highest brightness threshold T _h through a certain number of cloud-free images is to ensure the accuracy of clouds; calculating the lowest brightness threshold T _l through a certain number of cloud-containing images is to ensure the recall rate of clouds.

Step 3: Calculate the precise brightness threshold: analyze the image histogram in the initial image group and qualitatively screen cloud-free images; for cloud-containing images, use the roughly estimated brightness double threshold as the limiting condition to perform calculations based on the maximum inter-class variance to obtain the precise brightness threshold. .

Step 4: For the initial results after threshold segmentation, detect cloud areas with an area smaller than a ₁ , define them as highlight noise, delete them, and mark them as non-clouds; after deleting the highlight noise, perform a shape scale of a ₂ on the cloud-containing areas _The morphological expansion of Define it as a small cloud seam, delete it, and mark it as a cloud. The formula used for the above morphological expansion is:

In the formula, G is the brightness of the newly added pixel, is the brightness gradient in the expansion direction, d is a constant ranging from 0.05 to 0.25. _{The a1} , a2 and _a3 mentioned above are all configuration _parameters .

The final cloud content is obtained from the above steps.

S12 Detects multiple quality items on the images in the initial image group. The detection of quality items includes strip detection, high exposure detection, edge detection and histogram detection. The detection results of multiple quality items are normalized and the Values are normalized to mass fractions in the range 0-100.

S13 Integrate the cloud content and quality score to obtain the detection score. The numerical range of the detection score is 0-100.

In a specific embodiment of the present invention, step S2 includes:

S21 Sort the initial image group according to the detection score from high to low, and start filtering from the positive order, that is, start selecting the image with the highest detection score.

The selection principle is that adding the current image to the first image group can increase the coverage of the target area by the union of images in the first image group.

S22 Until the coverage of the target area by the union of the filtered images reaches the coverage threshold, the filtered images are used as the first image group. If the part of the target area covered by the current image is covered by the union of previously selected images, the current image is classified into the second image group.

Specifically, the coverage threshold is the maximum coverage of the target area by the union of filtered images, which can be 70%, 80% or 100%. In this embodiment, the coverage threshold is 80%.

Steps S21-S22 are explained according to a specific embodiment of the present invention: the 100 images in the initial image group A are sorted from high to low according to the detection scores, and the sorting is recorded as A = { A ₁ , A ₂ , .. ., A ₁₀₀ }, that is, A ₁ has the highest detection score. Starting from A ₁ , it is compared with the target area. If A ₁ covers part of the target area, A ₁ is classified as the first image group; then the subsequent images are Comparing with the target area respectively, the part of the target area covered by A ₃₅ is L ₁ , and this L ₁ has been completely covered by the 34 images { A ₁ , A ₂ ,..., A ₃₄ }, then A ₃₅ It was judged that it could not increase the coverage of the target area by the union of images in the first image group, and A ₃₅ was classified into the second image group. Finally, the first image group is recorded as B ={ B ₁ , B ₂ ,..., B _m }, the second image group is recorded as C ={ C ₁ , C ₂ ,..., C _n }, m + n =100, and m < n .

In a specific embodiment of the present invention, step S3 includes:

S31: Use a certain image in the first image group as the image to be expanded, intersect the image to be expanded with the union of the remaining images in the first image group except the image to be expanded, and divide the image to be expanded into are the intersecting area and the non-intersecting area, and the non-intersecting area is the area to be expanded.

S32 Sort the second image group according to the detection score from high to low, select the images in the second image group from the positive order to expand the area to be expanded, until the union of the selected images The coverage rate of the area to be expanded reaches 100%.

S33 Follow steps S31-S32 to expand all images in the first image group, and combine all images selected from the second image group with the first image group as a third image group.

Steps S31-S33 are explained according to a specific embodiment of the present invention: first , B 1 in the first image group B = { B ₁ , B ₂ ,..., B _m } is expanded _{, that is, B 1} is _to be When expanding the image, first determine the union U of { B ₂ , B ₃ ,..., B _m } images. The union U means { B ₂ , B ₃ ,..., B _m } - The image area shared between 1 image. Since the initial image group is an image filtered based on the target area, each image has the characteristics of large length, and there is a certain degree of regional intersection between the images, that is, B ₁ It also intersects with U to a certain extent, thereby dividing B ₁ into an intersecting area L ₂ and a non-intersecting area L ₃ , where L ₃ is the area to be expanded. After that, the second image group is sorted from high to low according to the detection score, and the images from the second image group C are selected to expand the area to be expanded L _3. The selection principle is also to start from the image with the highest detection score, and the image to be expanded. The areas are compared until the filtered image set C _i ¹ can completely cover the area L ₃ to be expanded. According to the above steps, { B ₂ , B ₃ ,..., B _m } are also expanded respectively, and the filtered image sets are { C _i ² , C _i ³ ,..., C _i ^m }, where the superscript 1,2,..., m represents the corresponding image set number obtained by expanding { B ₁ , B ₂ ,..., B _m }, and the subscript i represents the filtered image The number of images in the collection, i is an indefinite number, that is, the number represented by i in C _i ² and C _i ³ can be the same or different, i < n . Finally, { C _i ¹ , C _i ² , C _i ³ ,..., _C ^im } are deduplicated to obtain the extended image group C _i , C _i ⸦ C , and C _i and B are merged into the third image Group D.

In a specific embodiment of the present invention, step S4 includes:

Each image in the initial image group acquired by S41 includes metadata. The metadata includes the time of the image, remote sensor, region and other information. The information in the metadata is extracted and parsed to obtain time series data and sensor data.

S42 Calculate the variance of the time series data and the sensor data, normalize the quality score, the variance of the time series data, and the variance of the sensor data, and integrate them into a comprehensive score. The comprehensive score is The score range is 0-100.

S43 Construct the evaluation function based on the cloud content and the comprehensive score.

S44 Use the evaluation function to calculate the third image group to obtain the evaluation score of the third image group. The evaluation score is the normalized value of cloud cover and comprehensive score, recorded as E , and its value The range is still 0-100.

First, normalize the cloud content to a value between 0 and 100, recorded as X , and record the comprehensive score as Y. The formula for calculating the evaluation score is as follows:

Among them, ω ₁ is the weight coefficient of cloud content, ω ₂ is the weight coefficient of comprehensive score, ω ₁ > ω ₂ , and the sum of the two weight coefficients is 1. In this embodiment, ω ₁ is set to 0.7, and ω ₂ is set to 0.3.

S45 Sort the third image group from low to high according to its evaluation score, and filter from the positive order according to the screening strategy to obtain the preferred image group. The screening strategy is:

Step 1: Select a single image in sequence starting from the main sequence as the pre-screened image, and the image set in the third image group excluding the pre-screened image is used as the remaining image group;

Step 2: Set a retention rule. The retention rule is that if the coverage of the target area by the union of images in the remaining image group decreases after removing the pre-screened images, the corresponding pre-screened images will be directly removed. The filtered images are retained. If the pre-screened images do not meet the retention rules, the third step is executed;

Step 3: Calculate the average evaluation score of the third image group and the average evaluation score of the remaining image groups. If the average evaluation score of the remaining image group is lower than the average evaluation score of the third image group, then the corresponding pre-screened images will be reserve;

Step 4: If there are two or more images with the same evaluation scores among the images left after performing the third step, then the images with the same evaluation scores will be sorted from low to high cloud content. Keep the first image and filter out the remaining images.

Step S45 is explained according to a specific embodiment of the present invention: the third image group D is sorted according to the evaluation score from low to high and recorded as D = { D ₁ , D ₂ ,..., D _x }, starting from D ₁ Start executing the filtering strategy, for example, D ₁ is the pre-filtered image, { D ₂ ,..., D _x } is the remaining image group. First, D ₁ is judged according to the retention rules. If after removing D ₁ , the coverage of the target area by the union of images { D ₂ ,..., D _x } is reduced and is no longer 100%, then D ₁ is The image is determined to be necessary and retained; otherwise, it is still marked as pre-screened; then the average evaluation score of the third image group and the average evaluation score of the remaining image groups are calculated. For example, D ₃ does not meet the retention rules and is still marked as After pre-filtering the images, the remaining image group corresponding to D ₃ is { D ₁ , D ₂ , D ₄ ,..., D _x }. The remaining image group is recorded as K ³ and the average of all images in K ³ is calculated. The evaluation score is recorded as , and calculate the average evaluation score of all images in D , recorded as ,like , then D ₃ will be retained, otherwise, D ₃ will be screened out; after performing the above steps for all images in D , the retained images will be regarded as the retained image group, recorded as G = { G ₁ , G ₂ ,. .., G _y }, y < x , assuming that the evaluation scores of G _y _-4 , G _y _-2 and G _y are the same, then G _y _-4 , G _y _-2 and G _y will be ranked according to the cloud content from low to High sorting, the order after sorting is G _y _-2 < G _y < G _y _-4 . If G _y _-2 has the lowest cloud content, G _y _-2 will be retained, and G _y and G _y _-4 will be screened out. The images that were ultimately retained constitute the preferred image group.

In a specific embodiment of the present invention, step S5 includes:

S51: Use the images in the initial image group other than the preferred image group as candidate image groups.

S52 Determine the candidate image group based on the second expansion strategy, and the images that meet the requirements of the second expansion strategy are classified into the expanded image group.

S53 Use the expanded image group to expand the preferred image group to obtain the final image group.

The specific execution process of steps S52-S53 is as follows:

(1) Compare one image in the candidate image group with all the images in the preferred image group one by one,

(2) If there is no image in the preferred image group that covers 100% of the image in the candidate image group, then calculate the evaluation score of the image in the candidate image group. If it is higher than the If the average evaluation score of all images in the preferred image group is calculated, then the image in the candidate image group is classified into the extended image group;

(3) If there is an image in the preferred image group that covers 100% of the image in the candidate image group, then calculate the evaluation scores of the two images. If the image in the candidate image group If the evaluation score of the image is higher than the evaluation score of the image in the preferred image group, then the image in the candidate image group is classified into the extended image group;

(4) If there are two or more image groups in the preferred image group with a coverage rate of 100% for the image in the candidate image group, then compare the coverage of the image in the candidate image group. The evaluation score is the evaluation score of each image in the image group. If the evaluation score of the image in the candidate image group is higher than the lowest evaluation score in the image group, then the evaluation score of the image in the candidate image group is This image belongs to the extended image group.

(5) Combine the extended image group and the preferred image group into the final image group.

Steps S51-S53 are explained using a specific embodiment of the present invention: the preferred image group is recorded as Z = { Z ₁ , Z ₂ ,..., Z _s }, and the candidate image group is recorded as V = { V ₁ , V ₂ ,..., V _t }, Z + V = A , all images in V are compared with images in Z.

Taking V ₁ as an example, search in Z to see if there is an image that can completely cover V ₁ :

If no image in Z can completely cover V ₁ , then calculate the mean value of the evaluation scores of the s images { Z ₁ , Z ₂ ,..., Z _s } , and the evaluation score of V ₁ ,Compare and size, when , classify V ₁ into the extended image group.

If there is an image such as Z ₂ that can cover V ₁ , then calculate the evaluation scores of Z ₂ and V ₁ and ,when , classify V ₁ into the extended image group.

If there are 2 or more images in Z that can completely cover V ₁ , such as Z ₁ , Z _k , Z _l ( k , l < s ) all have 100% coverage of V ₁ , then Z is calculated first _1. Evaluation scores of Z _k and Z _l , sort the evaluation scores to get , and then calculate the evaluation score of V ₁ ,when , classify V ₁ into the extended image group.

The above steps are also performed for { V ₂ ,..., V _t } respectively to obtain the extended image group, and then the extended image group is added to the preferred image group to obtain the final image group.

The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present invention. All are covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

An accurate screening method for remote sensing image groups, including:

S1 selects a target area, obtains an initial image group within the target area, and calculates the detection score of the initial image group;

S2 divides the initial image group into a first image group and a second image group according to the detection score;

S3 expands the first image group based on the first expansion strategy to obtain a third image group;

S4 Calculate the evaluation score of the third image group, screen the third image group according to the evaluation score, and obtain the preferred image group;

S5 expands the preferred image group according to the second expansion strategy to obtain the final image group.
The precise screening method of remote sensing image groups according to claim 1, wherein step S1 includes:

Perform cloud cover detection on the images in the initial image group to obtain cloud content;

Perform quality item detection on the images in the initial image group to obtain a quality score, where the quality item detection includes strip detection, high exposure detection, edge detection and histogram detection;

The cloud content and the quality score are integrated to obtain a detection score, where the value range of the detection score is 0-100.
The precise screening method of remote sensing image groups according to claim 2, wherein step S2 includes:

Sort the initial image group from high to low according to the detection score, start filtering from the positive order, until the coverage of the union of the filtered images on the target area reaches the coverage threshold, and use the filtered image as the first image group, the images after excluding the first image group from the initial image group are used as the second image group.
The precise screening method of remote sensing image groups according to claim 2, wherein the quality score is obtained by integrating and calculating the detection results of each quality item, and the numerical range of the quality score is 0-100.
The precise screening method of remote sensing image groups according to claim 1, wherein step S3 includes:

Use one image in the first image group as the image to be expanded;

Determine the area to be expanded in the image to be expanded;

The second image group is sorted from high to low according to the detection score, and the images in the second image group are selected from the positive order to expand the area to be expanded until the selected second image group is The union of images reaches 100% coverage of the area to be expanded;

Traverse all the images in the first image group, and combine the selected images in the second image group with the first image group as a third image group.
The precise screening method of a remote sensing image group according to claim 5, wherein the image to be expanded intersects a union of the remaining images in the first image group except the image to be expanded, and the image to be expanded is divided into Intersecting area and non-intersecting area, the non-intersecting area is the area to be expanded.
The precise screening method of remote sensing image groups according to claim 2, wherein step S4 includes:

Obtain the metadata of the initial image group, and parse the metadata to obtain time series data and sensor data;

Calculate the variance of the time series data and the sensor data, and integrate the quality score, the variance of the time series data, and the variance of the sensor data into a comprehensive score;

Construct an evaluation function based on the cloud content and the comprehensive score;

Using the evaluation function to calculate the third image group, obtain an evaluation score of the third image group;

The third image group is sorted from low to high according to its evaluation score, and screened from positive order according to the screening strategy to obtain the preferred image group.
The precise screening method of remote sensing image groups according to claim 7, wherein the screening strategy is:

Step 1: Select a single image in sequence starting from the main sequence as a pre-screened image, and use the image set in the third image group excluding the pre-screened image as the remaining image group;

Step 2: Set a retention rule. The retention rule is that if the coverage of the target area by the union of the images in the remaining image group decreases after removing the pre-screened images, then the corresponding pre-screened images will be removed. The pre-screened images are retained. If the pre-screened images do not meet the retention rules, the third step is executed;

Step 3: Calculate the average evaluation score of the third image group and the average evaluation score of the remaining image group. If the average evaluation score of the remaining image group is lower than the average evaluation score of the third image group, then Keep the corresponding pre-screened images;

Step 4: If two or more images among the images retained after performing steps 2 and 3 have the same evaluation scores, the images with the same evaluation scores will be ranked in descending order of cloud content. Sort to highest, retain the first image, and filter out the remaining images.
The precise screening method of remote sensing image groups according to claim 1, wherein step S5 includes:

Use images in the initial image group other than the preferred image group as candidate image groups;

The candidate image group is determined based on the second expansion strategy, and images that meet the requirements of the second expansion strategy are classified into the expanded image group;

The preferred image group is expanded using the expanded image group to obtain a final image group.
The precise screening method of a remote sensing image group according to claim 9, wherein the candidate image group is determined based on the second expansion strategy, and the images that meet the requirements of the second expansion strategy are classified into the expanded image group. ,include:

Compare one image in the candidate image group with all the images in the preferred image group one by one,

If there is no image in the preferred image group that covers 100% of the image in the candidate image group, then calculate the evaluation score of the image in the candidate image group. If it is higher than the preferred image The average evaluation score of all images in the group, then the image in the candidate image group is classified into the extended image group;

If there is an image in the preferred image group that covers 100% of the image in the candidate image group, then the evaluation scores of the two images are calculated. If the evaluation of the image in the candidate image group If the score is higher than the evaluation score of the image in the preferred image group, then the image in the candidate image group is classified into the extended image group;

If there are two or more image groups in the preferred image group with a coverage rate of 100% for the image in the candidate image group, compare the evaluation score of the image in the candidate image group with The evaluation score of each image in the image group. If the evaluation score of the image in the candidate image group is higher than the lowest evaluation score in the image group, then the image in the candidate image group into the extended image group.