CN118296170A

CN118296170A - Warehouse entry preprocessing method and system for remote sensing images

Info

Publication number: CN118296170A
Application number: CN202410713852.9A
Authority: CN
Inventors: 陈宇; 陈婷; 明金; 吴皓; 任志宇; 安正雨; 杨丽帆; 万珍会; 程梦琦; 邹圣兵
Original assignee: Beijing Shuhui Spatiotemporal Information Technology Co ltd
Current assignee: Beijing Shuhui Spatiotemporal Information Technology Co ltd
Filing date: 2024-06-04
Publication date: 2024-07-05

Abstract

The invention provides a warehouse-in preprocessing method and a warehouse-in preprocessing system for remote sensing images, which relate to the technical field of data processing of remote sensing images, and specifically comprise the following steps: s1, acquiring a remote sensing image by adopting an image acquisition module; s2, detecting and analyzing the remote sensing image by using a data detection and analysis module to obtain a quality detection result, a quality score and an effective domain score of the image; s3, in an image screening module, screening the images according to the quality score and the effective domain score to obtain high-value images; s4, inputting the obtained high-value image and the quality detection result of the image into an image restoration module, and carrying out omnibearing restoration on the image problem area to obtain a restored image; s5, uploading the repaired image to a storage module for storage by adopting a data uploading module. The invention can automatically screen the images with higher use value when the remote sensing images enter the library, and repair the problems in all directions, thereby being convenient for the direct use of the subsequent images.

Description

Warehouse entry preprocessing method and system for remote sensing images

Technical Field

The invention relates to the technical field of remote sensing image processing and analysis, in particular to a warehouse entry preprocessing method and system for remote sensing images.

Background

With the rapid development of remote sensing technology, communication technology and computer technology, the earth observation field has entered into a big data era. The continuous increase of data types and data volume generates massive remote sensing image data of each level.

In the process of remote sensing image acquisition, due to sensor defects or the influence of an imaging mode and environment, some problem areas exist in the acquired remote sensing images, the usability of the images is influenced, after the acquired images are stored in a warehouse, related image files are called according to the requirements of users, and the problem areas in the images need to be repaired. Each time a user needs to repair the problem after retrieving and acquiring the required image, the complexity and the time consumption of the process are increased; because the data of the warehouse-in images are huge, the use frequency of some images is not high, and larger resource waste is caused if all the images are repaired, the development of the method is needed, the automatic repair of the images with higher use value in the warehouse-in process can be realized, and the complexity and the time consumption of the process in the subsequent use process are reduced.

Disclosure of Invention

The invention aims at providing a warehouse entry preprocessing method of remote sensing images aiming at the defects and shortcomings of the prior art. When the images are put into the warehouse, the method firstly carries out relevant calculation and analysis, screens and obtains a plurality of images with higher values, carries out omnibearing defect repair on the obtained high-value images to obtain repaired images, and then carries out warehouse storage, thereby being convenient for direct use when the images are subsequently called.

In order to achieve the above object, the present invention provides the following solutions: the invention provides a warehouse-in preprocessing method of remote sensing images, which comprises the following steps:

S1, acquiring a group of remote sensing images by adopting an image acquisition module, wherein the remote sensing images comprise a plurality of original images and corresponding metadata;

S2, detecting and analyzing the remote sensing images by using a data detection and analysis module to obtain quality detection results, quality scores and effective domain scores of the images;

S3, in an image screening module, screening a plurality of images according to the quality scores and the effective domain scores of the images to obtain high-value images;

S4, inputting the obtained high-value image and the quality detection result thereof into an image restoration module, and carrying out omnibearing restoration on a problem area of the high-value image to obtain a restored image;

And S5, uploading the repaired image to a storage module for storage by adopting a data uploading module.

The metadata in step S1 is description information of the remote sensing image, and includes: date and time of image acquisition, image acquisition range, image type, sensor, serial number, etc.

The data detection and analysis module in the step S2 comprises a first quality detection unit, a second quality detection unit and a data analysis unit;

Further, the first quality detection unit performs integrity detection on the obtained remote sensing image metadata, and analyzes the metadata detection condition based on an analysis method to obtain an original image with complete metadata; the second quality detection unit performs quality detection on the original image with complete metadata obtained in the first detection unit based on quality detection items to obtain quality detection results; and the data analysis unit performs data analysis according to the quality detection result to obtain the quality score and the effective domain score of the image.

In the step S2, the remote sensing images are detected and analyzed, and the specific steps of obtaining the quality detection result, the quality score and the effective domain score of each image are as follows:

S21, carrying out integrity detection on metadata of the remote sensing image through a first quality detection unit to obtain an original image with complete metadata;

S22, performing quality detection on the original image with complete metadata through a second quality detection unit to obtain a quality detection result of the image;

s23, analyzing and calculating the obtained quality detection result through a data analysis unit to obtain the quality score and the effective domain score of the image.

Further, the result obtained by the integrity detection of the metadata by the first quality detection unit in the step S21 includes three cases: complete metadata, metadata missing, metadata corrupted.

The original image with complete metadata enters a second quality detection unit for quality detection; the quality score and the available domain score of the metadata-missing and metadata-corrupted images are set directly to 0 score.

In the step S22, the specific steps of performing quality detection on the original image with complete metadata by using the second quality detection unit are as follows: and determining quality detection items of the original image, and respectively carrying out quality detection on the original image by using detection methods corresponding to the quality detection items to obtain detection results corresponding to the quality detection items.

Wherein the quality detection term comprises: side view angle detection, wave band matching detection, edge detection, null detection, cloud amount detection, shadow detection, histogram detection, stripe detection and high exposure detection, wherein the cloud amount detection comprises thin cloud detection and thick cloud detection, and the quality detection items can also comprise other detection items.

The quality detection results include three cases: defect-free, slight defect, severe defect.

The second quality detection unit is configured with a plurality of quality detection item detection methods, and can detect the full quality detection item of the remote sensing image, and can detect the quality detection item required by the remote sensing image, and the detection method of each quality detection item can be multiple.

Further, in the step S23, the specific process of analyzing and calculating the quality score and the effective domain score of the image by the data analysis unit is as follows:

S231, obtaining the quality score of the image and the defect area of the image in each quality detection item according to the detection results of all quality detection items;

the defective areas of the images in the quality detection items are areas where defects exist in the images in the quality detection items, and the defective areas comprise slightly defective areas and severely defective areas.

The quality score is obtained by setting score values of different buckles according to the defect degree of detection results of all detection items through percentage calculation, and the specific calculation rules are as follows:

Wherein i represents the detection item category, and S _i represents the corresponding score value of the detection result of the detection item category i. The failure rating is preferentially subtracted as the false withholding of the severe failure.

S232, the defect areas of the images in the quality detection items are obtained in a union mode, the problem areas of the images are obtained, areas except the problem areas in the images are obtained to serve as effective areas of the images, and the effective area fraction of the images is obtained through calculation according to the effective areas.

The method for calculating the effective domain score comprises the following steps:

Effective domain score = effective region size/image size x 100.

In step S3, the high-value image is an image with an effective domain not smaller than a first threshold and a quality score not smaller than a second threshold.

The problem area in the step S4 includes: edge anomaly regions, null regions, thick cloud regions, thin cloud regions, fog regions, shadow regions, histogram anomaly regions, band regions, and high exposure regions.

The image restoration module in the step S4 includes a classification unit, a first restoration unit, a second restoration unit, a third restoration unit and a quality inspection unit.

The first repair unit repairs the first quality problems, and the second repair unit repairs the second quality problems; the third repairing unit is used for repairing third quality problems; the classifying unit classifies the quality detection result of the input image according to the repairing method to obtain a first type quality problem, a second type quality problem and a third type quality problem.

The specific process of performing the omnibearing repairing on the problem area of the high-value image in the step S4 is as follows:

S41, dividing the problem area into a first class of quality problems, a second class of quality problems and a third class of quality problems according to different restoration methods based on the defects of the high-value image problem area in the classification unit;

S42, repairing the first, second and third quality problems by a first, second and third repairing units by adopting a certain repairing method to obtain a first type repairing area, a second type repairing area and a third type repairing area;

S43, quality inspection is carried out on the first type of repair area and the second type of repair area through a quality inspection unit, so that the first type of repair area and the second type of repair area which are qualified in quality inspection, the first type of repair area and the second type of repair area which are unqualified in quality inspection are obtained;

S44, repairing the first type of repairing area with unqualified quality inspection and the second type of repairing area with unqualified quality inspection in a third repairing unit again, and merging the first type of repairing area and the second type of repairing area into the third type of repairing area;

s45, forming a repaired area by the qualified first type of repair area, the qualified second type of repair area and the third type of repair area, and taking the repaired area as a repair result of the problem area of the image to be repaired, thereby obtaining the repaired image.

Further, the repair method in the step S42 includes a first repair method, a second repair method, and a third repair method.

The first repairing unit adopts a first repairing method to repair the first quality problems to obtain a first repairing area; the second repairing unit adopts a second repairing method to repair the second class quality problem to obtain a second repairing area; and the third repairing unit adopts a third repairing method to repair the third type of quality problems, the first type of repairing areas with unqualified quality inspection and the second type of repairing areas with unqualified quality inspection, so as to obtain the third repairing areas.

The first repairing method carries out information complementation on the image missing part through the information of the image, realizes the repairing of the first quality-capable problem under lower cost, and ensures the information authenticity of the repairing area to a certain extent.

The second repairing method adopts an image enhancement scheme to enhance the contrast and the details of the image, so that the second quality problem is repaired, and the information authenticity of the repaired area is ensured.

And the third repairing method adopts an image filling method to fill and repair the third type of quality problems, the first type of repairing areas with unqualified quality inspection and the second type of repairing areas with unqualified quality inspection.

Further, in the step S43, the specific process of performing quality inspection on the first type of repair area and the second type of repair area by the quality inspection unit is as follows:

S431, taking the first type of repair area and the second type of repair area as areas to be inspected, and performing image conversion on the areas to be inspected to obtain gray images of the first type of repair area and the second type of repair area;

S432, dividing the obtained gray image into square image blocks with a certain size;

s433, calculating and evaluating the homogeneity degree of each image block;

S434, taking the image blocks of the region to be inspected, the image blocks of which have the degree of homogenization greater than a homogenization threshold value, as qualified image blocks;

s435, taking the region to be inspected, of which the ratio of the qualified image blocks to the images of all the regions to be inspected is larger than the set qualification rate threshold value, as a repair region of qualified quality inspection.

The invention also provides a remote sensing image warehousing pretreatment system applied to the remote sensing image warehousing pretreatment method, which specifically comprises the following steps: the system comprises an image acquisition module, a data detection and analysis module, an image screening module, an image restoration module and a data uploading module; the data detection and analysis module comprises a first quality detection unit, a second quality detection unit and a data analysis unit; the image restoration module comprises a classification unit, a first restoration unit, a second restoration unit, a third restoration unit and a quality inspection unit.

The beneficial effects of the invention are as follows: the invention constructs an evaluation calculation method of the effective domain data and the quality data of the remote sensing image, and the method is adopted to evaluate the effective domain and the quality when the remote sensing image is put in storage, screen out the image with higher available value, carry out omnibearing restoration on the problems of the screened image, store in storage, directly retrieve when the image is used later, and do not need to carry out restoration treatment, thereby improving the use and retrieval efficiency of the image.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flow chart of a warehouse entry preprocessing method for remote sensing images provided by the invention;

Fig. 2 is a schematic structural diagram of a warehouse entry preprocessing system for remote sensing images provided by the invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which are derived by a person skilled in the art based on the embodiments of the invention, fall within the scope of protection of the invention.

Fig. 1 is a flowchart of a method for preprocessing a remote sensing image in storage, and the method of the present invention is described in further detail below for each step in the embodiment flow, and specifically includes the following steps:

the metadata is description information of the remote sensing image, describes parameters related to remote sensing image acquisition, and comprises the following steps: date and time of image acquisition, image range, image type, sensor, serial number, etc.

The data detection and analysis module comprises a first quality detection unit, a second quality detection unit and a data analysis unit;

And the first quality detection unit is mainly used for carrying out integrity detection on the obtained remote sensing image metadata, and analyzing the metadata detection condition based on an analysis method to obtain an original image with complete metadata.

The result obtained by carrying out integrity detection on the metadata comprises three cases: complete metadata, metadata missing, metadata corrupted.

The metadata is complete, which means that the file of the metadata can be opened normally, the content of the metadata is not missing, the file name of the metadata is matched with the content of the file, and the like. The case of metadata deletion includes content incomplete of metadata, file name incomplete of metadata, information incomplete in a file of metadata, and the like. The case of metadata corruption includes failure of a file of metadata to open, file format errors of metadata, and the like.

The original image with complete metadata enters a second quality detection unit for quality detection, and the quality score and the available domain score of the original image with missing metadata and damaged metadata are directly set to 0 score.

The second quality detection unit performs quality detection on the original image with complete metadata based on quality detection items to obtain quality detection results;

The specific steps for carrying out quality detection on the original image with complete metadata based on the quality detection item to obtain a quality detection result are as follows:

determining quality detection items of the original image, and respectively carrying out quality detection on the original image by using detection methods corresponding to the quality detection items to obtain detection results corresponding to the quality detection items;

The second quality detection unit is configured with a plurality of quality detection item detection methods, and can detect the full quality detection item of the remote sensing image, and can detect the quality detection item required by the remote sensing image, and the detection method of each quality detection item can be multiple. For example, the cloud amount detection method may be a physical threshold method, a full-probability bayesian method, or other detection methods, where the cloud amount detection result is the cloud amount of the remote sensing image.

And the data analysis unit performs data analysis according to the detection result obtained by the second quality detection unit to obtain the quality score and the effective domain score of the image.

The specific process for obtaining the quality score and the effective domain score according to the quality detection result comprises the following steps:

1. obtaining the quality score of the image and the defect area of the image in each quality detection item according to the detection results of all quality detection items;

the defective area of the image in each quality detection item is the area where the image has defects in each quality detection item.

The quality scores are calculated by setting different deduction values according to the defect degree of the detection result of each detection item, and the specific calculation rules are as follows:

Wherein i represents the detection item category, and S _i represents the detection result deduction value of the detection item category i. The failure rating is preferentially subtracted as the false withholding of the severe failure.

The setting of the score value of each quality test item is as follows:

(1) Side view detection

In order to shorten the coverage period of satellite images on the ground and improve the acquisition capability of satellite data, the existing high-resolution satellites have the side-view imaging capability, and the side-view imaging or the phenomena of space resolution degradation, positioning capability reduction and the like are caused, and along with the increase of the side view angle, the image point error caused by topography fluctuation is increased, so that the side view angle has a direct influence on the usability of remote sensing images.

In the side view angle detection, the range [0 DEG, 90 DEG ] of the side view angle is divided into a plurality of side view angle number ranges, each number range corresponds to a deduction value, and the image quality is deducted according to the detection result of the side view angle.

The number range and the corresponding point number set in this embodiment are shown in the following table:

side view angle number range	Defect grade	Score value
			[0°,5°]	Defect free	0
(5°,10°]	Minor defects	5
			(10°,20°]	Minor defects	10
(20°,90°]	Serious defect	100

(2) Band matching detection

The band matching refers to whether the number of the bands of the remote sensing image meets the requirement or not and whether the bands are missing or not. The specific scoring process of the band matching detection comprises the following steps: reading the number of full-color spectrum bands and the number of multispectral bands of the remote sensing image, wherein if the band numbers are matched, the remote sensing image is free of defects, and the deduction value is 0; if the band numbers are not matched, the defect is serious, and the deduction value is 100; and deducting the image quality according to the detection result of the band matching.

(3) Edge detection

The edge detection mainly identifies abnormal edges in the image by extracting and evaluating boundary information between a target object and a background in the image. The specific scoring process of edge detection is as follows: if the edge detection result is abnormal, no edge defect exists, and the deduction value is 0; if the edge color is abnormal, the defect is slight, and the deduction value is 5; and buckling the image quality according to the detection result of the edge.

(4) Null value detection

Performing null detection on the remote sensing image by using a pixel brightness detection method to obtain a null pixel area, and performing null detection score evaluation according to the proportion of the null area to the whole image; in the null value detection, the score evaluation considers the size of the image connected domain influenced by the null value in addition to the null value duty ratio, so that a minimum connected domain threshold R is set to combine the null value duty ratio and simultaneously perform the null value detection score evaluation. The specific scoring process of null value detection is as follows: if the image does not meet the requirement of the minimum connected domain threshold R, the detection term deduction value is 100; if the requirement of the minimum connected domain threshold R is met, dividing the null value duty ratio range [0%,100% ] into a plurality of null value number domains, wherein each number domain corresponds to a deduction value, and deducting the image quality according to the detection result of the null value.

The null value range and the corresponding point setting conditions set in this embodiment are shown in the following table:

Null value number range	Defect grade	Score value
			[0%,1%]	Defect free	0
(1%,5%]	Minor defects	5
			(5%,20%]	Minor defects	10
(20%,50%]	Minor defects	20
			(50%,100%]	Minor defects	40
Minimum connected domain < R	Serious defect	100

(5) Cloud detection

Cloud quantity is defined as the proportion of a cloud area (comprising thin cloud and thick cloud) in an image to the whole image, and cloud quantity detection score evaluation is carried out according to the proportion of the cloud area to the whole image; in cloud amount detection, the score evaluation considers the cloud amount duty ratio and also considers the size of an image connected domain influenced by the cloud, so that a minimum connected domain threshold R is set to combine the cloud amount duty ratio and simultaneously perform the cloud amount detection score evaluation. The specific scoring process of cloud cover detection comprises the following steps: if the image does not meet the requirement of the minimum connected domain threshold R, the detection term deduction value is 100; if the requirement of the minimum connected domain threshold R is met, the cloud amount range [0%,100% ] is divided into a plurality of cloud amount value domains, each value domain corresponds to a deduction value, and the image quality is deducted according to the cloud amount detection result.

The cloud amount value range and the corresponding deduction value setting conditions set in the embodiment are shown in the following table:

Cloud amount value range	Defect grade	Score value
			[0%,5%]	Defect free	0
(5%,10%]	Minor defects	2.5
			(10%,30%]	Minor defects	10
(30%,60%]	Minor defects	20
			(60%,100%]	Minor defects	40
Minimum connected domain < R	Serious defect	100

(6) Shadow detection

And obtaining a shadow region of the image by adopting a shadow detection method based on HSV color space, and evaluating a shadow detection score according to the proportion of the shadow region to the whole image. The specific scoring process of shadow detection is as follows: dividing the shadow duty ratio range [0%,100% ] into a plurality of shadow number value fields, wherein each number value field corresponds to a deduction value, and deducting the image quality according to the shadow detection result.

The setting conditions of the shadow number value fields and the corresponding deduction values set in the embodiment are shown in the following table:

Shadow numerical field	Defect grade	Score value
			[0%,5%]	Defect free	0
(5%,20%]	Minor defects	5
			(20%,30%]	Minor defects	10
(30%,60%]	Minor defects	20
			(60%,100%]	Minor defects	40

(7) Histogram detection

Histogram detection utilizes the statistical properties of the remote sensing image to identify potential anomalies or quality problems. The brightness value of each pixel of the remote sensing image can form a histogram, and the abnormal region in the image can be rapidly found by analyzing the distribution characteristics of the histogram. The specific scoring process of the histogram detection is as follows: detecting that the histogram abnormality does not exist, and setting the deduction value to be 0 if the histogram abnormality does not exist; if there is a histogram abnormality, the defect is serious, and the score value is set to 100.

The histogram abnormality is a detection result after removing cloud, shadow and high exposure influence. For hyperspectral image histogram detection, a multidimensional histogram is built based on histogram characteristics of different wave bands, abnormal pixels are determined by calculating a mahalanobis distance and the like, and for the histogram abnormality of a specific wave band, slight defects are set, and each wave band is buckled by 2 minutes.

(8) Strip detection

The stripe detection is mainly used for identifying and positioning stripe noise in an image, the stripe noise is caused by various factors such as sensor faults, data transmission problems or atmospheric conditions, and the stripe noise is usually represented as a horizontal, vertical or oblique bright or dark stripe in the image, so that the quality of a remote sensing image can be seriously influenced, and the accuracy of the remote sensing image in applications such as ground object classification, target identification and change detection is reduced. The stripe detection includes a row stripe detection method and a column stripe detection method. The specific scoring process for strip detection is as follows: if no stripe problem is detected, no stripe defect is detected, and the deduction value is set to be 0; if the problem of the strip exists, the defect is slight, and the deduction value is set to be 10; and buckling the image quality according to the strip detection result.

(9) High exposure detection

A high exposure is a situation where a pixel or region typically exhibits a significantly higher brightness than other regions. The high exposure is carried out by detecting the brightness value of each regional pixel of the remote sensing image, screening to obtain pixel blocks with brightness value larger than 250, calculating the ratio of the total area of all the pixel blocks with brightness larger than 250 to the total area of the whole remote sensing image, and carrying out high exposure detection fraction evaluation according to the ratio. The specific scoring process of the high exposure detection is as follows: the high exposure duty ratio range [0%,100% ] is divided into a plurality of high exposure number value fields, each number value field corresponds to a deduction value, and the image quality is deducted according to the detection result of the high exposure.

The setting conditions of the high exposure value range and the corresponding deduction value set in this embodiment are shown in the following table:

High exposure value range	Defect grade	Score value
			0%	Defect free	0
(0%,5%]	Minor defects	5
			(5%,10%]	Minor defects	10
(10%,20%]	Minor defects	20
			(20%,100%]	Minor defects	40

2. And taking the union set of the defect areas of the images in the quality detection items to obtain problem areas of the images, obtaining areas except the problem areas in the images as effective areas of the images, and calculating the effective domain score of the images according to the effective areas.

Effective domain score = effective region size/image size x 100.

S3, in an image screening module, screening the images according to the obtained effective domain score and the quality score to obtain high-value images;

the high-value image is: the effective domain is not smaller than the first threshold value, and the quality score is not smaller than the image of the second threshold value.

In this embodiment, the first threshold and the second threshold are set to 80.

S4, inputting the obtained high-value image and the quality detection result thereof into an image restoration module, and carrying out omnibearing restoration on a problem area of the image to obtain a restored image;

the problem area includes: edge anomaly regions, null regions, thick cloud regions, thin cloud regions, fog regions, shadow regions, histogram anomaly regions, band regions, and high exposure regions.

The image restoration module comprises a classification unit, a first restoration unit, a second restoration unit, a third restoration unit and a quality inspection unit.

The classifying unit classifies the defect problems in the quality detection result of the input image according to the repairing method to obtain a first type quality problem, a second type quality problem and a third type quality problem.

Setting the first type of quality problem in this example includes: an edge anomaly region, a null region, and a stripe region; the second category of quality problems includes: a thin cloud region, a fog region, a shadow region, and a histogram abnormal region; the third category of quality problems includes: thick cloud areas.

The first repair unit repairs the first quality problems, and the second repair module repairs the second quality problems; and the third repairing module is used for repairing third class quality problems.

The specific process for carrying out the omnibearing restoration on the problem area of the high-value image comprises the following steps:

s41, dividing the problem area into a first class of quality problems, a second class of quality problems and a third class of quality problems according to different restoration methods based on the defects of the high-value image problem area in the classification unit.

S42, repairing the first, second and third quality problems by a certain repairing method through the first, second and third repairing units to obtain a first type repairing area, a second type repairing area and a third type repairing area.

The repair methods include a first repair method, a second repair method, and a third repair method.

The first repairing unit adopts a first repairing method to repair the first quality problems to obtain a first repairing area; the second repairing unit adopts a second repairing method to repair the second class quality problem to obtain a second repairing area; and the third repairing unit adopts a third repairing method to repair the third quality problems to obtain a third repairing area.

The first repairing method carries out information complementation on the missing part through the information of the image, realizes the repairing of the first quality problem at lower cost, and ensures the information authenticity of the repairing area to a certain extent.

In this embodiment, a linear interpolation method is used to repair the first quality problem, and the specific steps are as follows:

1. Binarizing the image area of the first quality problem, wherein 0 represents a normal pixel position, and 1 represents a pixel position needing to be repaired;

2. For each pixel to be repaired, the known normal data pixels around it are selected as reference points. Typically, four pixels are selected, up, down, left and right, closest to the pixel to be repaired, and for high-dimensional data (e.g., multispectral, hyperspectral images), adjacent pixels in the corresponding band are selected;

if the pixel to be repaired is positioned at the edge of the image, when the number of available pixels is small, the searching range can be properly enlarged, for example, more adjacent pixels are selected inwards along the edge, or the reference points are increased by considering the use of edge expansion technology (such as mirroring and period prolongation);

3. Based on the selected reference points, constructing a one-dimensional or two-dimensional (determined according to the data dimension) linear model to fit the points to obtain a linear model, and calculating the slope and intercept according to the coordinates of the reference points and the corresponding gray values (or spectrum values);

For the one-dimensional case, assuming that the reference points are (x ₁,y₁) and (x ₂,y₂), and the position of the pixel to be repaired is x, the linear model is y=mx+b, wherein m is the slope, b is the intercept, and the slope m and the intercept b are calculated according to the reference points to obtain the linear model;

4. Performing interpolation calculation on each pixel point in the missing or abnormal region by using the constructed linear model, namely substituting the position of the pixel to be repaired into the linear model to obtain a pixel estimated value of the pixel;

5. filling the pixel value obtained by interpolation calculation into a missing or abnormal region of the original image to finish the repair of the pixel;

6. Repeating the steps for all pixels marked as needing to be repaired until all the missing or abnormal pixels are subjected to linear interpolation to obtain pixel values and are filled back into the original image, and performing smoothing treatment on the repaired image to obtain a first type of repair area.

The first repairing method can also adopt polynomial interpolation, spline interpolation, inverse distance weight interpolation, kerling interpolation and other methods.

The second repairing method enhances the contrast of the image through image enhancement, further enhances the details of the image, realizes the repairing of the second quality problem, and ensures the information authenticity of the repairing area.

In this embodiment, a frequency domain enhancement repair method is used to repair the second class quality problem, and the specific steps are as follows:

1. converting the second quality problem area from a space domain to a frequency domain by adopting fast Fourier transformation to obtain high-frequency information and low-frequency information of the second quality problem area, wherein the low-frequency part corresponds to a large-scale structure and slow change characteristics of the image, and the high-frequency part corresponds to details and edges of the image;

The specific process of obtaining the high-frequency information and the low-frequency information of the second quality problem area by adopting the fast Fourier transform is as follows:

(1) The second type of quality problem area f (x, y) is expressed as the product of the illumination component i (x, y) and the reflection component r (x, y), i.e.:

Where i (x, y) represents the illumination component, representing information describing the illumination of the scene, which changes slowly and can be considered as the low frequency part of the image; r (x, y) represents the reflected component, information representing scene details, which changes rapidly and can be seen as the high frequency part of the image.

(2) Carrying out logarithmic operation on f (x, y), separating out related components of i (x, y) and r (x, y), and then carrying out Fourier transformation and frequency domain processing on the components to obtain high-frequency information and low-frequency information of a second class quality problem area, namely:

Where DFT denotes a fourier transform function, ln denotes a logarithmic function, F _i (u, v) denotes low-frequency information of the second-type quality problem area, and F _r (u, v) denotes high-frequency information of the second-type quality problem area.

2. High-pass filtering is carried out on the high-frequency information and the low-frequency information which are subjected to Fourier transformation and frequency domain processing, the low-frequency information is weakened, the high-frequency information is enhanced, and a filtering result is obtained, namely:

Where S (u, v) represents the filtering result and H (u, v) represents the high-pass filtering function.

3. Converting the processed frequency domain image back to a spatial domain by adopting inverse discrete Fourier transform, and obtaining a repaired image by utilizing exponential operation, thus obtaining a second repair area;

Where g (x, y) represents the second type of repair area, i ₀ (x, y) represents the illumination component of the second type of repair area, and r ₀ (x, y) represents the incident component of the second type of repair area.

And the third repairing method adopts an image filling method to fill and repair the third type of quality problems, the first type of repairing areas with unqualified quality inspection and the second type of repairing areas with unqualified quality inspection, so as to obtain the third type of repairing areas.

In this embodiment, a sample-based repair method is used to repair the third quality problem, and the specific steps are as follows:

1. taking the third quality problem area as a target area, and taking the image to be repaired with the third quality problem area as a target image;

2. searching and screening a reference image for repairing a target image, wherein the reference image comprises a target area, and the target area has no quality defect;

The reference image should be as similar as possible to the target image in terms of content, viewing angle, lighting conditions, etc. The content includes characteristics of color, texture, shape, etc.

3. Accurately aligning the reference image and the target restoration image by using an image registration technology (such as feature point matching, optical flow estimation, affine transformation, perspective transformation and the like) so as to ensure that the corresponding positions of the target area and the reference image content are consistent;

4. In the registered reference image, selecting a complete, clear and content-related area corresponding to the position of a target area in the target image as a filling area, cutting out a required filling area, and performing operations such as size adjustment, rotation, mirror image overturning and the like according to the requirement to ensure that the required filling area is perfectly matched with the target area of the target image;

5. and splicing the cut filling area of the reference image with the target area of the target image to carry out filling and repairing to obtain a third type of repairing area.

For complex scenes, a mixing strategy, such as transparency mixing (Alpha blending), poisson blending (Poisson blending) and other methods, can be adopted, so that the filling area and the original content of the target image are naturally transited, and the visual splitting sense is reduced.

S43, quality inspection is carried out on the first type of repair area and the second type of repair area through the quality inspection unit, and the first type of repair area and the second type of repair area which are qualified in quality inspection, and the first type of repair area and the second type of repair area which are unqualified in quality inspection are obtained.

The specific process and standard of the quality inspection unit for the first type of repair area and the second type of repair area are as follows:

1. taking the first type of repair area and the second type of repair area as areas to be inspected, performing image conversion on the areas to be inspected to obtain gray images of the first type of repair area and the second type of repair area, and marking the gray images of the areas to be inspected as R;

2. the resulting gray-scale image is divided into square image blocks of size mxm, each centered on a pixel (x, y), the image blocks being represented as:

Wherein K _x,y represents an image block centered on a pixel (x, y), f (i, j) represents a pixel value of a pixel point of K _x,y in the image block, Ω _x,y represents a neighborhood centered on (x, y), and the image size is mxm;

3. the degree of homogeneity of each image block is calculated and evaluated, and the degree of homogeneity of the image block K _x,y is expressed as:

Wherein, Representing the degree of homogeneity of image block K _x,y, |k _x,y | represents the number of pixels in image block K _x,y, and Φ (i, j) represents the local variation of pixel point (i, j).

The local change is calculated according to the pixel values of the pixel point (i, j) and the surrounding four adjacent domain pixels, and the specific mode is as follows:

。

the homogeneity degree reflects the difference of the intensities of adjacent pixels, and the smaller the difference is, the larger the homogeneity degree of the image is.

And setting a homogenization threshold value T _h, judging whether the image block of the region to be inspected is qualified or not according to the homogenization degree H _K of the image block and the homogenization threshold value T _h, and marking as qualified if H _K>Τ_h and marking as unqualified if H _K≤Τ_h.

Judging whether the quality inspection of the first type of repair area and the second type of repair area is qualified or not according to the ratio of the qualified image blocks to the images of all the areas to be inspected and the set threshold value.

Since not all local areas of the image are suitable for evaluating the image quality by using the homogeneity, firstly, the image block K _x,y of the area to be detected is screened by the image block suitable for homogeneity detection, and the specific process is as follows:

1. calculating pixel value variances in each image block K _x,y:

Wherein, Representing pixel value variances of pixels in the image block K _x,y; The average value of pixel values representing pixels in each image block K _x,y is calculated as follows:

2. Selecting an image block set S _r to be screened, and calculating to obtain the variance of the image block set S _r to be screened:

Wherein S _r represents a set of image blocks to be screened, which accounts for 15% of the total number of image blocks of the region to be inspected, |S _r | represents the number of image blocks in the image block S _r, Representing the variance of the set of image blocks S _r to be screened.

3. And screening according to the difference value between the variance of the image block set S _r to be screened and the variance of each image block K _x,y to obtain an image block set S _e for calculating the homogenization degree:

Wherein S _e represents a filtered image block set S _e;T_σ capable of calculating a degree of homogeneity, and a threshold of a difference between a variance of the image block set S _r to be filtered and a variance of each image block K _x,y.

S44, repairing the first type of repairing area with unqualified quality inspection and the second type of repairing area with unqualified quality inspection again in a third repairing unit by adopting a third repairing method, and merging the repairing areas into the third type of repairing area.

Fig. 2 is a schematic structural diagram of a remote sensing image warehousing pretreatment system according to the present invention, which further describes the remote sensing image warehousing pretreatment system according to the present invention in detail, specifically including: the system comprises an image acquisition module, a data detection and analysis module, an image screening module, an image restoration module and a data uploading module; the data detection and analysis module comprises a first quality detection unit, a second quality detection unit and a data analysis unit; the image restoration module comprises a classification unit, a first restoration unit, a second restoration unit, a third restoration unit and a quality inspection unit.

The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any changes or substitutions easily contemplated by those skilled in the art within the scope of the present invention should be included in the present invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.

Claims

1. The warehousing pretreatment method for the remote sensing image is characterized by comprising the following steps of:

s2, detecting and analyzing the remote sensing images by adopting a data detection and analysis module to obtain quality detection results, quality scores and effective domain scores of the images;

S3, in an image screening module, screening a plurality of images according to the quality score and the effective domain score of each image to obtain high-value images;

2. The method of claim 1, wherein the data detection and analysis module in step S2 includes a first quality detection unit, a second quality detection unit, and a data analysis unit.

3. The method for preprocessing the remote sensing image according to claim 2, wherein the specific steps of detecting and analyzing the remote sensing image in the step S2 to obtain the quality detection result, the quality score and the effective domain score of each image are as follows:

4. The method for pre-processing a remote sensing image according to claim 3, wherein the step S22 of performing quality detection on the original image with complete metadata by using the second quality detection unit comprises the following specific steps: determining quality detection items of the original image, and performing quality detection on the original image by using a detection method corresponding to each quality detection item to obtain a detection result corresponding to each quality detection item;

The second quality detection unit is configured with a plurality of quality detection item detection methods, and can detect a plurality of quality detection items on the original image, and the detection method of each quality detection item can be multiple.

5. The method for preprocessing a remote sensing image in storage according to claim 4, wherein in step S23, the quality score and the effective domain score of the image are obtained by analysis and calculation by the data analysis unit, specifically comprising:

S232, the defect areas of the images in the quality detection items are obtained in a union mode, a problem area of the images is obtained, areas except the problem area in the images are obtained to serve as effective areas of the images, and the effective area fraction of the images is obtained through calculation according to the effective areas;

Effective domain score = effective region size/image size x 100.

6. The method according to claim 5, wherein the defect area of the image in each quality inspection item in step S231 is the area of the image with the defect in each quality inspection item; the quality scores are obtained by setting different deduction values according to the defect degree of the detection result of each detection item and adopting percentage calculation, and the specific calculation rules are as follows:

Wherein i represents the category of the detection item, and S _i represents the score value of the detection result corresponding to the category i of the detection item.

7. The method for preprocessing a remote sensing image according to claim 1, wherein the high-value image in the step S3 is an image with an effective domain not smaller than a first threshold and a quality score not smaller than a second threshold.

8. The method of claim 1, wherein the image restoration module in step S4 includes a classification unit, a first restoration unit, a second restoration unit, a third restoration unit, and a quality inspection unit.

9. The method for preprocessing the remote sensing image in storage according to claim 8, wherein the step S4 of performing omnibearing restoration on the problem area of the high-value image comprises the following specific steps:

S41, dividing the problem area into a first class of quality problems, a second class of quality problems and a third class of quality problems according to different repairing methods based on defects of the high-value image problem area in the classifying unit;

s42, repairing the first, second and third quality problems by adopting corresponding repairing methods through the first, second and third repairing units to obtain a first type repairing area, a second type repairing area and a third type repairing area;

10. The method for preprocessing a remote sensing image according to claim 9, wherein the repairing method in step S42 includes a first repairing method, a second repairing method and a third repairing method; the first repairing unit adopts a first repairing method to repair the first quality problems to obtain a first repairing area; the second repairing unit adopts a second repairing method to repair the second class quality problem to obtain a second repairing area; the third repairing unit adopts a third repairing method to repair a third type of quality problem, a first type of repairing area with unqualified quality inspection and a second type of repairing area with unqualified quality inspection to obtain a third repairing area;

the first repairing method carries out information complementation on the image missing part through the information of the image, realizes the repairing of the first quality problem under low cost, and ensures the information authenticity of the repairing area to a certain extent;

the second repairing method adopts an image enhancement scheme, and realizes the repairing of the second quality problem by enhancing the image contrast and the image detail, thereby ensuring the information authenticity of the repaired area;

11. The method for pre-processing the remote sensing image according to claim 9, wherein the specific process of performing quality inspection on the first type of repair area and the second type of repair area by the quality inspection unit in step S43 is as follows:

s433, calculating and evaluating the homogeneity degree of each image block;

12. A remote sensing image warehouse entry preprocessing system, characterized in that the system is applied to the method of any one of the above claims 1-11, and the system specifically comprises: the system comprises an image acquisition module, a data detection and analysis module, an image screening module, an image restoration module, a data uploading module and a storage module;