A remote sensing image frame loss rapid detection method

Abstract

The application discloses a kind of remote sensing image frame loss quick detection method, belong to remote sensing image quality evaluation and processing field.It includes: remote sensing image is carried out equal proportion downsampling, generates remote sensing image thumbnail, detects and extracts the dark pixel area of remote sensing image thumbnail;Determine whether the image exists image edge area frame loss condition;Mask image is generated, and mask image is carried out morphological operation;According to the mask image after morphological operation processing, extract image connected domain, obtain the effective area of image;Determine whether the image exists image middle area frame loss condition.The application can quickly and accurately detect whether remote sensing image exists frame loss problem, and does not need to additionally make frame loss image sample data set, simple and easy to operate, effectively avoids the false detection of frame loss image itself characteristics susceptible to special scene interference, realizes the automatic quality evaluation of remote sensing image.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image quality assessment and processing, and in particular to a rapid method for detecting frame loss in remote sensing images, applicable to application scenarios such as remote sensing image quality assessment, frame loss detection, and product manufacturing. Background Technology

[0002] In the production and application of remote sensing image products, the quality assessment of remote sensing images is crucial. Pre-evaluation through qualitative or quantitative assessments allows for the selection of high-quality images for subsequent product production and application release, improving user experience and production efficiency. Reducing the need for secondary screening of low-quality image data significantly saves manpower, resources, and time.

[0003] There are two main traditional methods for assessing the quality of remote sensing images: one is manual judgment, the other is to build a mathematical model based on image statistical information, and the third is to identify dropped frames based on convolutional neural networks to assess the quality.

[0004] The first method, relying on subjective evaluation through manual judgment, yields the best results. However, this method requires significant manpower and time, the evaluation process is cumbersome, and it heavily depends on prior human knowledge, resulting in low efficiency and limited applicability. Given the current explosive growth in remote sensing data, relying on manual judgment for quality assessment is impractical.

[0005] The second method mainly uses statistical information such as image signal-to-noise ratio, variance, mean, and correlation coefficient to assess the quality of remote sensing images. However, this method is easily affected by scene information, and its stability and adaptability are insufficient. It can only make simple evaluations of the images and cannot accurately identify specific quality problems in the images, such as frame loss.

[0006] The third method trains anomaly data using convolutional neural networks to identify dropped frames. However, training a high-accuracy identification model requires creating a large dataset of dropped frame samples, resulting in high time and financial costs.

[0007] In summary, none of the above methods can achieve low-cost and rapid frame loss detection for massive amounts of remote sensing images. Moreover, they all simply design algorithm models based on the characteristics of frame loss itself, and only judge the appearance of frame loss. It is difficult to systematically integrate the variable phenomena of frame loss corresponding to massive and complex images into the algorithm model. Summary of the Invention

[0008] The purpose of this invention is to provide a rapid method for detecting frame drops in remote sensing images, avoiding the problems existing in the background art described above. This invention is simple, fast, and effective, requiring no additional data collection of frame-dropped image samples. It can quickly and accurately detect whether frame drops exist in remote sensing images, effectively avoiding false detections caused by the inherent characteristics of frame-dropped images being susceptible to interference from special scenes, thus achieving automated quality assessment of remote sensing images. Furthermore, it has a wide range of applications, suitable for horizontal frame drop zones, vertical frame drop zones, multi-angle oblique frame drop zones, and special frame drop scenarios appearing in image edge regions.

[0009] The technical solution adopted by this invention to solve its technical problem is as follows:

[0010] A method for rapid detection of frame loss in remote sensing images includes the following steps:

[0011] (1) Based on the image size, the remote sensing image is downsampled proportionally to generate a remote sensing image thumbnail, and the dark pixel region of the remote sensing image thumbnail is detected and extracted.

[0012] (2) Determine whether there is frame dropping in the image edge area. If so, output the detection result and end the detection. Otherwise, continue with the subsequent steps.

[0013] (3) Generate a mask image, perform morphological operations on the mask image, and obtain the mask image after morphological operation processing;

[0014] (4) Extract connected components of the image based on the mask image after morphological operation and obtain the effective region of the image;

[0015] (5) Determine whether there are frame drops in the middle region of the image, output the detection results, and complete the fast frame drop detection.

[0016] Furthermore, the specific method of step 1 is as follows:

[0017] (101) Acquire a remote sensing image I, whose length and width are denoted as W and H, respectively;

[0018] (102) Downsample the remote sensing image I proportionally to generate a thumbnail with a length and width no greater than 1024. If W≤1024 or H≤1024, no action is taken; otherwise, calculate the downsampling ratio:

[0019]

[0020] `max` represents the maximum value.

[0021] (103) Extract thumbnails All dark pixels, where a dark pixel is a pixel that satisfies the following condition:

[0022]

[0023] In the formula, thumbnail width, thumbnail height, thumbnail The cell value in the j-th row and i-th column;

[0024] (104) Thumbnail Connected dark pixels in the image form a dark pixel region A. i Get thumbnails All dark pixel regions A i(i＝1,2,...,M) M is a thumbnail The number of mid-dark pixel regions.

[0025] Furthermore, the specific method for step 2 is as follows:

[0026] (201) Traverse thumbnails All dark pixel regions in the candidate set of dropped images are added to the candidate set V when the length w and width h of the dark pixel region meet the following conditions:

[0027] or

[0028] Where T is a set threshold, T≥0.98;

[0029] (202) Count the total number of pixels N in each dark pixel region in the candidate set V of the dropped frame image. pixels As long as there exists a dark pixel region, the total number of pixels N pixels Satisfying N pixels If the value is ≥0.8*W*H, the image is determined to have a frame dropping problem, and the frame dropping occurs in the edge area of ​​the image. The detection ends; otherwise, continue to step 3.

[0030] Furthermore, step 3 is performed as follows:

[0031] (301) Generating and Thumbnails Mask image of the same size Mask image Cell value in row j and column i for:

[0032]

[0033] (302) Mask image Perform morphological closing operations, and eliminate the mask image by first dilating and then eroding. Fine black dot noise in the middle, smoothing the boundaries of dark pixel regions; expansion and erosion radius R = 3;

[0034] (303) The mask image after morphological operation processing Recorded as

[0035] Furthermore, step 4 is specifically implemented as follows:

[0036] (401) Processed mask image Extracting image connected component C j,j＝1,2,3...,N N is the number of connected components;

[0037] (402) Statistical analysis of each connected component C in the image j,j＝1,2,3...,N The four boundaries of X min,j X max,j Y min,j Y max,j ; where X min,j X max,j Y min,j Y max,j These correspond to the leftmost column index, rightmost column index, bottom row index, and top row index of the j-th connected component in the image, respectively.

[0038] (403) Define the connected component C j,j＝1,2,3...,N The four vertices of P i,j,i＝1,2,3,4 Among them, P 1,j P 2,j P 3,j P 4,j Connected components C j,j＝1,2,3...,N The topmost point, rightmost point, bottommost point, and leftmost point, that is:

[0039] P 1,j .y=Y max,j

[0040] X min,j <P 1,j .x <X max,j

[0041] P 2,j x = X min,j

[0042] Y min,j <P 2,j .y <Y max,j

[0043] P 3,j .y=Y min,j

[0044] X min,j <P 3,j.x <X max,j

[0045] P 4,j x = X max,j

[0046] Y min,j <P 4,j .y <Y max,j

[0047] Where .x represents the x-component and .y represents the y-component;

[0048] (404) Connected component C that satisfies the following two conditions j Defined as the effective area R of the image:

[0049] ① Connected region C j The four vertices P 1,j P 2,j P 3,j P 4,j There are 4 distinct vertices;

[0050] ② Connected region C j It is approximately a parallelogram.

[0051] Furthermore, step 5 is specifically implemented as follows:

[0052] The number of valid regions S in the image is counted. If S≥2, then the image has a frame dropping problem, and the frame dropping occurs in the middle region of the image. The detection result is output to complete the fast frame dropping detection.

[0053] The beneficial effects of this invention are as follows:

[0054] 1. This invention can quickly and accurately detect frame loss in remote sensing images without the need to create additional sample datasets of frame loss images. It is simple and easy to implement, effectively avoiding false detections caused by the inherent characteristics of frame loss images being susceptible to interference from special scenes, and realizing automated quality assessment of remote sensing images.

[0055] 2. This invention has a wide range of applications, and can be applied to horizontal frame loss bands, vertical frame loss bands, multi-angle oblique frame loss bands, and special frame loss scenarios that appear in the edge area of ​​the image.

[0056] 3. This invention provides a new method for detecting frame loss in remote sensing images, which is highly efficient. Attached Figure Description

[0057] Figure 1 This is a flowchart of a method for rapid detection of frame loss in remote sensing images provided by an embodiment of the present invention.

[0058] Figure 2 , Figure 3This is a typical example of frame loss in remote sensing imagery where frame loss occurs in the central region of the image.

[0059] Figure 4 This is an example image of a remote sensing image with dropped frames appearing in the edge area of ​​the image.

[0060] Figure 5 This is an example image of a non-frame-loss remote sensing image that is prone to false detection during frame loss detection.

[0061] Figure 6 yes Figure 2-5 A schematic diagram of the corresponding effective area of ​​the remote sensing image. Detailed Implementation

[0062] The present invention will now be further described with reference to the accompanying drawings.

[0063] A fast method for detecting frame drops in remote sensing images, such as Figure 1 As shown, it includes the following steps:

[0064] (1) Based on the image size, the remote sensing image is downsampled proportionally to generate a remote sensing image thumbnail, and the dark pixel regions of the remote sensing image thumbnail are detected and extracted; the specific method is as follows:

[0065] (101) Acquire a remote sensing image I, whose length and width are denoted as W and H, respectively;

[0066] (102) Downsample the remote sensing image I proportionally to generate a thumbnail with a length and width no greater than 1024. If W≤1024 or H≤1024, no action is taken; otherwise, calculate the downsampling ratio: `max` represents the maximum value.

[0067] (103) Select thumbnail Perform frame drop detection and extract All dark pixels, where a dark pixel is a pixel that satisfies the following condition:

[0068]

[0069] In the formula, thumbnail width, thumbnail height, thumbnail The cell value in the j-th row and i-th column;

[0070] (2) Determine whether there is frame dropping in the image edge area. If so, output the detection result and end the detection. Otherwise, continue with the subsequent steps.

[0071] Frame loss in remote sensing images can be systematically categorized into two situations: one is that frame loss occurs in the central region of the image, and the image is divided into multiple regions by the frame loss area; the other is that frame loss occurs in the edge region of the image, and the image is cropped into one or more regions.

[0072] The specific method for step 2 is as follows:

[0073] ① Traverse thumbnails All dark pixel regions, when dark pixel region A i A frame is added to the candidate set V of dropped images when its length w and width h satisfy the following conditions:

[0074]

[0075] Where T is a set threshold, generally T≥0.98;

[0076] ②Statistically analyze the dark pixel regions A in the candidate set V of the dropped image. i Total number of pixels N pixels When there is a dark pixel region with a total number of pixels N pixels When the following conditions are met: N pixels If the image size is ≥0.8*W*H, then the image has a frame dropping problem, and the frame dropping occurs in the edge area of ​​the image.

[0077] This step effectively eliminates the areas with no values ​​corresponding to the four corner points of the corrected satellite image, as shown in the attached image. Figure 5 As shown;

[0078] (3) Generate a mask image, perform morphological operations on the mask image to obtain the morphologically processed mask image; the specific method is as follows:

[0079] (301) Generating and Thumbnails Mask image of the same size Mask image Cell value in row j and column i for:

[0080]

[0081] (302) Mask image Perform morphological closing operations, and eliminate the mask image by first dilating and then eroding. Fine black dot noise in the middle, smoothing the boundaries of dark pixel regions; expansion and erosion radius R = 3;

[0082] (303) The mask image after morphological operation processing Recorded as

[0083] (4) Extract connected components from the mask image after morphological operations and obtain the effective region of the image; the specific method is as follows:

[0084] (401) Mask image after morphological operation processing Use Python's findContours function to extract connected components C from an image. j,j＝1,2,3...,N N is the number of connected components in the image;

[0085] (402) Statistical analysis of each connected component C in the image j,j＝1,2,3...,N The four boundaries of X min,j X max,j Y min,j Y max,j ; where X min,j X max,j Y min,j Y max,j These correspond to the leftmost column index, rightmost column index, bottom row index, and top row index of the j-th connected component in the image, respectively.

[0086] (403) Define the connected component C j,j＝1,2,3...,N The four vertices of P i,j,i＝1,2,3,4 Among them, P 1,j P 2,j P 3,j P 4,j Connected components C j,j＝1,2,3...,N The topmost point, rightmost point, bottommost point, and leftmost point, that is:

[0087] P 1,j .y=Y max,j

[0088] P 2,j x = X min,j

[0089] P 3,j .y=Y min,j

[0090] P 4,j x = X max,j

[0091] (404) Define the connected region C that satisfies the following two conditions as the effective region R of the image:

[0092] ① The connected region C has four distinct vertices P1, P2, P3, and P4, that is:

[0093] P 1,j ≠P 2,j And P 1,j ≠P 4,j

[0094] P 3,j ≠P 2,j And P 3,j ≠P 4,j

[0095] In detail, generally as shown in the appendix Figure 2 and attached Figure 3 ,have:

[0096] P1.y=Y max X min <P1.x<X max

[0097] P2.x=X min Y min <P2.y<Y max

[0098] P3.y=Y min X min <P3.x<X max

[0099] P4.x=X max Y min <P4.y<Y max

[0100] For special circumstances, see attached Figure 4 Then we have:

[0101] P1.y=Y max P1.x = X min

[0102] P2.x=X min P2.y = Y max

[0103] P3.y=Y min P3.x = X max

[0104] P4.x=X max P4..y=Y min

[0105] ② The connected domain C is roughly a parallelogram, that is, the lengths of the corresponding sides of the four vertices P1, P2, P3, and P4 are approximately the same; in order to avoid interference from circles or other shapes, the coordinates of the four vertices and the number of pixels in the connected domain are statistically analyzed, and the two are combined to determine whether it is an approximate parallelogram.

[0106] Figure 6 for Figures 2-5 A schematic diagram of the effective area of ​​remote sensing imagery under certain conditions.

[0107] (5) Count the number of valid regions S in the image. If S≥2, it means that the image is divided into multiple regions by the frame loss region. The image has a frame loss problem. Output the detection result and complete the fast frame loss detection.

[0108] In summary, this invention can quickly and accurately detect frame loss issues in remote sensing images without requiring the creation of additional frame loss image sample datasets. It is simple and easy to implement, effectively avoiding false detections caused by the inherent characteristics of frame loss images being susceptible to interference from special scene conditions, thus achieving automated quality assessment of remote sensing images. This invention has a wide range of applications, suitable for horizontal frame loss bands, vertical frame loss bands, multi-angle oblique frame loss bands, and special frame loss scenarios appearing in image edge regions.

[0109] Finally, it should be noted that the above embodiments are merely illustrative examples to illustrate the spirit of this invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, but without departing from the spirit of this invention or exceeding the scope defined by the appended claims.

Claims

1. A method for rapid detection of frame drops in remote sensing images, characterized in that, Includes the following steps: (1) Based on the image size, the remote sensing image is downsampled proportionally to generate a remote sensing image thumbnail, and the dark pixel regions of the remote sensing image thumbnail are detected and extracted; the specific method is as follows: (101) Acquire a remote sensing image The length and width of the image are denoted as follows: and ; (102) Remote sensing images Proportional downsampling generates thumbnails with a length and width no greater than 1024. ,like or If the result is positive, no action is taken; otherwise, calculate the downsampling ratio factor: ； `max` represents the maximum value. (103) Extract thumbnails All dark pixels, where a dark pixel is a pixel that satisfies the following condition: In the formula, thumbnail Length, thumbnail width, thumbnail In the Line number The cell values ​​of the column; (104) Thumbnail Connected dark pixels in a dark pixel region form a dark pixel region. Get thumbnails All dark pixel regions M is a thumbnail The number of mid-dark pixel regions; (2) Determine whether there are frame drops in the image edge region. If so, output the detection result and end the detection; otherwise, continue with the subsequent steps. The specific method is as follows: (201) Traverse thumbnails All dark pixel regions, when the length of the dark pixel region Hekuan The dark pixel region is added to the candidate set of dropped images when the following conditions are met. : in, To set a threshold, ; (202) Statistical candidate set of dropped images Total number of pixels in each dark pixel region As long as there exists a dark pixel region, the total number of pixels satisfy If the frame drop occurs, the image is determined to have a frame drop problem, and the frame drop occurs in the edge area of ​​the image, and the detection ends; otherwise, continue to step 3. (3) Generate a mask image, perform morphological operations on the mask image to obtain the morphologically processed mask image; the specific method is as follows: (301) Generation and thumbnails Mask image of the same size Mask image In the Line number Column cell values for: (302) Mask image Perform morphological closing operations, and eliminate the mask image by first dilating and then eroding. Fine black dot noise, smoothing the boundaries of dark pixel regions; dilation and erosion radius is 3; (303) The mask image after morphological operation processing Recorded as ; (4) Extract connected components of the image based on the morphologically processed mask image, and obtain the effective region of the image; the specific method is as follows: (401) The processed mask image Extracting connected components from images , The number of connected components; (402) Statistical analysis of each connected component in the image The four boundaries , , , ;in, , , , Corresponding to the first in the image The leftmost column index value, rightmost column index value, bottom row index value, and top row index value of each connected component; (403) Define connected components The four vertices are ;in, , , , Connected components The topmost point, rightmost point, bottommost point, and leftmost point, that is: in, Represents the x-component. Represents the y-component; (404) Connected components that satisfy the following two conditions Defined as the effective area of ​​the image : ① Connected components The four vertices , , , There are 4 distinct vertices; ② Connected domain It is a parallelogram; (5) Determine whether there are frame drops in the central region of the image, output the detection results, and complete the fast frame drop detection; the specific method is as follows: Count the number of valid regions in the statistical image ,like If the image has a frame drop problem and the frame drop occurs in the middle region of the image, the detection result is output to complete the fast frame drop detection.

Images