Black edge detection and removal method and device for remote sensing image and medium

Abstract

The invention discloses a black edge detection and removal method and device for a remote sensing image, and a medium. The method comprises the following steps: carrying out strip division preprocessing on the remote sensing image; a multi-index parallel strategy is adopted to count the feature statistical value of each strip; using a bidirectional mutation detection algorithm to compare the characteristic statistical values of the adjacent strips, and obtaining a black edge suspected area; after structure verification is carried out on the suspected black-edge area, multi-scale strip fusion analysis is carried out to locate a black-edge area; and carrying out geometric characteristic constraint on the accurate positioning area, dividing a black edge area and carrying out labeling to generate a structured report. According to the method, various regular and incomplete black edges such as image edge strip-shaped black edges, center strip-shaped black edges and edge plane-shaped black edges can be detected, the defect that most of existing algorithms can only aim at one or two of the black edges is overcome, meanwhile, multi-sensor data adaptation is adopted, the method is suitable for various remote sensing data types, and the method is suitable for large-scale popularization and application. The limitation that a traditional quality inspection system is only suitable for a single sensor source is broken through.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent processing of remote sensing images and computer vision, and in particular to a black edge detection and removal method for remote sensing images, a device and a medium. The present application is particularly aimed at the problem of invalid data of black edges commonly existing in remote sensing images obtained by satellite remote sensing (including multi-source data such as optical, SAR, hyperspectral, etc.), unmanned aerial vehicle aerial photography, and aerial photogrammetry platforms, and provides a data quality inspection processing solution. BACKGROUND

[0002] With the development of remote sensing technology, remote sensing images have a wide range of applications in various geographic information systems (GIS), environmental monitoring, and land use planning. However, during the processing of remote sensing images, due to the quality of the remote sensing data itself or the processing of the mosaicking line editing step, black edges (i.e. the outer periphery of the image that has not been collected or invalid area) are generated, which is an important factor affecting the image quality and subsequent processing accuracy.

[0003] In remote sensing image processing, map production, and image quality inspection applications, automatic detection of image edge black edges is a basic and critical preprocessing task. In the prior art, there are mainly the following several black edge detection methods, but all have certain limitations, for example:

[0004] Statistical method based on the proportion of pure black pixels. This method usually sets a fixed gray threshold (such as a gray value less than 10), and by counting the proportion of low gray pixels in the edge region of the image, it is determined whether there is a black edge. This method is simple to implement, but in the case of gradual change, grayish, compression artifacts, or changes in shooting conditions (such as backlight, dark field acquisition), it is easy to miss detection, and when the overall image brightness is dark, it is easy to produce false positives, and its robustness is poor.

[0005] Edge change detection method based on fixed threshold difference. By setting a fixed gray jump threshold, it is detected whether there is a sudden change in the edge gray value to determine the presence of a black edge. Due to the fixed threshold, it is difficult to adapt to different sensors, different image types, and shooting environments, especially when there are high-contrast edge objects (such as white frames, superimposed text, icons, etc.), false positives are easy to occur, and its reliability is insufficient.

[0006] Method based on edge operator (such as Canny, Sobel, etc.) to extract edge features. This type of method works well for sharp edges, but its effectiveness decreases significantly when dealing with non-continuous black edges, smooth gradual black edges, or blurred image edges. At the same time, it is easily disturbed by complex background textures, producing a large number of false edges, resulting in insufficient detection accuracy.

[0007] Black edge detection methods based on deep learning models. This kind of method automatically learns black edge features by training a convolutional neural network model, and has certain adaptive ability. However, the deep learning method generally relies on a large number of high-quality labeled samples for training, and the model migration has the problem of insufficient generalization ability when adapting to new sensors or new resolution images, and the training and deployment process requires high computing resources, and there is certain difficulty in engineering application.

[0008] For example: the invention application with document number CN113344937A discloses a method for automatically identifying and removing black edges of remote sensing images. The method can automatically identify and remove redundant black edges of remote sensing images, and is suitable for geometrically corrected remote sensing images. However, the scheme has the problems of low processing efficiency for ultra-large size remote sensing images, poor effect in processing complex background and irregular shape black edges, and low recognition accuracy. In addition, the method depends on specific preprocessing steps, such as geometric correction, and its applicability is limited for remote sensing images without or with poor preprocessing effect.

[0009] In summary, the existing black edge detection methods all have the problems of insufficient detection ability for gradual black edges and weak boundary black edges, poor robustness to background interference, poor adaptability, and high deployment cost, and an automatic black edge detection method that does not require a large amount of training data, can consider gradual black edges and complex boundary characteristics, has strong adaptability and high robustness is urgently needed. SUMMARY

[0010] In view of the above problems, the purpose of the present application is to provide a black edge detection and removal method, device and medium for remote sensing images, which solves the problems of insufficient detection ability for gradual black edges and weak boundary black edges, poor robustness to background interference, poor adaptability and high deployment cost in the prior art.

[0011] The present application provides a black edge detection and removal method, device and medium for remote sensing images.

[0012] The first aspect is a black edge detection and removal method for remote sensing images, comprising:

[0013] S1, performing strip division preprocessing on the remote sensing image;

[0014] S2, using a multi-index parallel strategy to count the feature statistical value of each strip;

[0015] S3, using a bidirectional mutation detection algorithm to compare the feature statistical values of adjacent strips to obtain a black edge suspected area;

[0016] S4, after structure verification of the black edge suspected area, performing multi-scale strip fusion analysis to locate the black edge area;

[0017] S5, geometric property constraint is performed on the accurate positioning area, a black border area is divided and labeled, and a structured report is generated.

[0018] Optionally, the remote sensing image categories include optical images, SAR images, and infrared images, wherein the optical images are divided into RGB images, multispectral images, or hyperspectral images.

[0019] The strip division width is automatically adjusted according to the remote sensing image resolution, and the formula is:

[0020]

[0021] wherein, is the strip width, is the remote sensing image height, is the rounding function.

[0022] Optionally, in S2, the multi-index parallel strategy includes the mean (μ) and the standard deviation (σ) of the RGB or grayscale channel.

[0023] wherein, the multi-index feature statistics value is updated in real time by using a sliding window algorithm, and a strip feature matrix is established to store the multi-index statistics value of each strip.

[0024] Optionally, in S3, a bidirectional mutation detection algorithm is used, and when mutation detection is performed, an adaptive method based on global statistics is used to dynamically determine a comparison threshold, including the following steps:

[0025] S31, the overall distribution of all strip multi-index feature values is calculated.

[0026] S32, the median of the difference between adjacent strip feature values is taken as a reference threshold T.

[0027] S33, the reference threshold value is multiplied by an adjustment coefficient α according to the type of the remote sensing image, and a continuous mutation number threshold N is set.

[0028] wherein, the bidirectional mutation detection algorithm is represented by the following formula:

[0029] or

[0030] wherein, is the mean μ threshold, is the standard deviation σ threshold; if the mutation is detected for N times continuously in the mutation area, a black border area warning is triggered.

[0031] Optionally, in S4, the black border suspected area is subjected to structure verification, including:

[0032] Extract the Sobel edge feature map of the image, and perform structure verification according to the edge density in the black edge early warning area, the texture complexity and the brightness component V of the HSV color space.

[0033] Optionally, the multi-scale strip fusion analysis in S4 is used to locate the black edge area, and the multi-scale strip fusion analysis comprises the following steps:

[0034] S41, preliminary screening and positioning of a suspected area is performed by using a wide-scale strip;

[0035] S42, accurate positioning is performed by using a fine-scale strip in the suspected range;

[0036] S43, the detection results of different scales are fused through a voting mechanism to determine the black edge area.

[0037] Optionally, the geometric property constraint on the accurate positioning area comprises the following steps:

[0038] Discrete detection points are connected through morphological closing operation to filter isolated noise areas;

[0039] For a non-rectangular image, minimum circumscribed rectangle calibration is performed first, and then strip division is performed.

[0040] Optionally, the structured report in S5 comprises the following steps:

[0041] The black edge area ratio BAD is calculated, and the formula is as follows:

[0042]

[0043] The quality grading result is output according to the black edge area ratio BAD, and the quality grading result comprises the following steps:

[0044] I is BAD<1%, II is 1%≤BAD<5%, III is (BAD≥5%), and a visual result layer is additionally provided, wherein the black edge area is marked by a semi-transparent red mask.

[0045] The second aspect: an electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor executes the program to realize the steps of the method provided in the first aspect.

[0046] The third aspect: a non-transitory computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the steps of the method provided in the first aspect.

[0047] The beneficial effects of the present application are as follows:

[0048] 1. The present application can detect various shapes of incomplete black edges such as edge strip-shaped black edges, image center strip-shaped black edges, and image edge planar black edges, which makes up for the shortcomings of existing algorithms that can only detect one or two types of black edges. The present application also uses multi-sensor data adaptation, which is suitable for optical images (RGB / multi-spectral / high-spectral), SAR images, infrared images, and other types of remote sensing data. It breaks through the limitations of traditional quality inspection systems that are only suitable for single sensor sources and has the ability to quickly analyze large-scale remote sensing data.

[0049] 2. The present application divides the image into multiple strips by rows or columns and uses a sliding window algorithm to update the statistical values in real time, greatly improving the calculation efficiency. It can be applied to the processing of ultra-large images or high-bit-depth images, solving the problem of large calculation amount in traditional convolution algorithms based on image edges when processing ultra-large size remote sensing images. The method integrates GPU acceleration, online threshold learning, and other intelligent technologies, supporting real-time processing from satellite data to offline batch inspection in all scenarios, further improving the processing speed.

[0050] 3. The present application automatically adjusts the strip width according to the image resolution, making the detection method better adapt to remote sensing images of different resolutions. It automatically identifies the number of bands in the input image, calculates the statistical features of each band, and then performs weighted fusion to ensure effective processing of multi-spectral images. It also provides a strict mode (suitable for surveying and mapping level products) and a relaxed mode (suitable for rapid screening) dual threshold strategy, which can be flexibly selected according to different application scenarios and needs, enhancing the adaptability of the system.

[0051] 4. The present application calculates the mean, standard deviation, median, and range of RGB or grayscale channels to provide comprehensive features for black edge detection, improving recognition accuracy. It uses an adaptive method to determine the dynamic threshold and sets a continuous mutation number threshold to avoid misjudgment and automatically adjust the threshold. The bidirectional mutation detection algorithm synchronously compares the statistical values of both ends of the image and judges the black edge by combining the continuous mutation number. It uses a second difference method to enhance the sensitivity of gradual black edges. Edge structure verification and multi-scale fusion further improve detection accuracy, reduce misjudgment, and improve result reliability.

[0052] 5. The present application method is easy to apply in engineering and solves the problem that existing quality inspection algorithms mostly stop at the level of mathematical model and algorithm research, which is not convenient for engineering application. It can also generate a structured report, mark the black edge area coordinate range, calculate the black edge area ratio, and output the quality classification result, with additional visual result layers marked by a semi-transparent red mask to facilitate manual review and quality evaluation in engineering applications. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1A flowchart of the black edge detection and removal method of the present application is shown in Figure 1.

[0054] Figure 2 A schematic diagram of the large black edge of the embodiment of the present application is shown in Figure 2.

[0055] Figure 3 A schematic diagram of the rectangular image black edge of the embodiment of the present application is shown in Figure 3.

[0056] Figure 4 A structural diagram of the black edge detection and removal device of the present application is shown in Figure 4.

[0057] Figure 5 A structural diagram of the electronic device of the present application is shown in Figure 5. DETAILED DESCRIPTION

[0058] The embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar symbols represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as a limitation of the present application.

[0059] The existing remote sensing image black edge detection has the problems of small algorithm detection coverage, not suitable for large-scale remote sensing data rapid quality inspection and analysis, less research on model sub-item parameter configuration and reconstruction method, and inconvenient for engineering application and promotion.

[0060] In view of the above problems, the present application provides a black edge detection and removal method for remote sensing images, Figure 1 A flowchart of the black edge detection and removal method for remote sensing images provided by the embodiment of the present application is shown in Figure 1, which comprises:

[0061] S1, strip division preprocessing is performed on the remote sensing image.

[0062] The remote sensing image category includes optical images, SAR images and infrared images, etc., wherein the optical images are further divided into RGB images, multispectral images or hyperspectral images.

[0063] The SAR image is an active microwave remote sensing image, and its black edge characteristics are different from those of the optical image. When processing, its radar backscatter characteristics need to be considered, and the black edge is detected by analyzing the intensity value, texture and other characteristics of the image. For example, the backscatter intensity of the black edge area is usually low, and the texture is uniform.

[0064] The infrared image reflects the thermal radiation characteristics of the ground object, and the temperature characteristics of the black edge area may be different from those of the normal area. The black edge is detected by analyzing the temperature distribution, thermal texture and other characteristics of the infrared image. For example, the black edge area may exhibit low temperature and uniform distribution.

[0065] RGB images utilize information from their R, G, and B color channels to perform preliminary black border identification based on color characteristics. For example, in the RGB color space, the channel values ​​of black border areas are usually low. By setting a threshold range for the average value of each channel, suspected black border areas can be preliminarily screened out.

[0066] Multispectral images contain multiple bands (such as blue, green, red, and near-infrared), and each band is processed separately. By fusing information from multiple bands, the difference between black borders and normal areas is enhanced, improving the accuracy of black border detection. For example, certain bands may be more sensitive to black borders; combining information from these bands for analysis can more accurately identify black borders.

[0067] Hyperspectral imagery possesses extremely fine spectral resolution (typically ranging from tens to hundreds of bands), allowing for more precise segmentation and identification of black borders using its rich spectral information. By analyzing the spectral characteristics of the black border region in different bands and comparing them with the spectral characteristics of normal ground features, the extent of the black border can be accurately defined.

[0068] In remote sensing image processing, optical images play a crucial role. This invention further refines the processing strategy for optical images. Specifically, RGB images, multispectral images, and hyperspectral images are treated separately for black border detection due to their differences in spectral information and resolution. RGB images, with their rich color information, are suitable for preliminary black border identification based on color features; multispectral images, through the fusion of information from multiple bands, enhance the accuracy of black border detection; and hyperspectral images, with their fine spectral resolution, enable more precise segmentation and identification of black borders. This differentiated processing strategy for different types of optical images significantly improves the efficiency and accuracy of black border detection and removal.

[0069] like Figure 2 and Figure 3 As shown, the remote sensing image data to be inspected is read and converted into a computable digital matrix form. To improve inspection efficiency, a striping strategy is adopted: for horizontal inspection mode, the image is divided into several horizontal strips of equal height along the row direction (e.g., one strip for every 10 rows); for vertical inspection, vertical strips of equal width are divided along the columns. During segmentation, the consistency of strip width must be maintained, while considering any residual rows / columns that may exist at the image boundaries. Full coverage is ensured by dynamically adjusting the width of the last strip.

[0070] The strip width is determined using a dynamic strip width algorithm, which automatically adjusts the strip width based on the image resolution. The formula is as follows:

[0071]

[0072] in, For the strip width, For the height of the remote sensing image, is an integer function.

[0073] Using the above division method, the device can efficiently process remote sensing image data. Through the built-in dynamic strip width algorithm module, the device can automatically adjust the strip width according to the resolution of the remote sensing image, ensuring that the image is reasonably and efficiently divided. This flexible division method not only improves the detection efficiency, but also ensures that the residual rows / columns of the image boundary can be covered by the last strip after dynamic adjustment, thereby avoiding the omission of information. Using the above division method, the device lays the foundation for the subsequent black edge detection and removal steps.

[0074] S2, adopt multi-index parallel strategy to calculate the statistical value of each strip feature.

[0075] The statistical feature calculation of each strip adopts a multi-index parallel strategy: in addition to calculating the mean (μ) and standard deviation (σ) of the RGB or grayscale channel, it can also supplement the calculation of robustness indicators such as median and range. For multispectral images, the statistical quantities of each band need to be calculated separately.

[0076] For example, for a horizontal strip of the R channel, the mean value is:

[0077]

[0078] where m is the strip height, n is the strip width, is the R channel pixel value at that position.

[0079] The standard deviation (σ) can be used to calculate the dispersion of each channel pixel value relative to the mean value, and the standard deviation can be expressed as:

[0080]

[0081] where, is the pixel value.

[0082] Further, to improve calculation efficiency, a sliding window algorithm is used to update the statistical values in real time, for example, a suitable size of sliding window (such as a 3x3 window) is set, and it is moved pixel by pixel within the strip. At the same time, a strip feature matrix is established to store the μ, σ and other parameters of each strip, providing a complete data basis for subsequent abrupt change analysis. When the multi-index feature statistical values are updated in real time using the sliding window algorithm, a strip feature matrix is established to store the multi-index statistical values of each strip.

[0083] S3, adopt a bidirectional abrupt change detection algorithm to compare the statistical values of adjacent strips to obtain suspected black edge regions.

[0084] Before mutation detection, a dynamic threshold adaptive mechanism is used to dynamically determine the baseline threshold. The adaptive method based on global statistics can be used to calculate the overall distribution of all bands μ and σ in the whole image, and the median of the difference between adjacent bands is taken as the baseline threshold. Then, according to the image type (such as optical / SAR), multiply by the adjustment coefficient α (typical value 1.5-3.0). At the same time, set the continuous mutation number threshold N (usually 3-5 bands) to avoid false judgment caused by single noise.

[0085] Specifically, the statistical quantities of adjacent bands are compared synchronously from both ends to the center of the image for mutation region detection, and the formula is expressed as:

[0086] Or

[0087] Wherein, is the mean μ threshold, is the standard deviation σ threshold; if the mutation is detected for N times in a region, the black edge region warning is triggered.

[0088] Using the above method, the statistical quantities of adjacent bands are compared synchronously from both ends to the center of the image, and when the judgment condition is met, the mutation event is recorded. If the mutation is detected for N times in a region, the black edge warning is triggered.

[0089] Further, for gradual black edges (such as feathered edges), a second difference method is used to enhance the change sensitivity.

[0090] The second difference method can more finely capture the small changes of the edge, and can effectively identify even if the edge transition is relatively smooth. The second difference method can significantly improve the accuracy of black edge detection, ensure that even gradual black edges can be accurately captured and warned, and provide strong support for subsequent black edge removal work.

[0091] S4, after structure verification of the black edge suspected area, multi-scale band fusion analysis is performed to locate the black edge region.

[0092] After preliminary detection, structure verification is performed on the suspected area: first, the Sobel edge feature map of the image is extracted, the edge density in the black edge warning area is analyzed, and the real black edge usually shows a sudden drop in edge density (ΔED < β). The contrast of the warning area is calculated by the gray level co-occurrence matrix (GLCM) to verify the texture complexity. The texture complexity of the real black edge region is significantly lower than that of the normal region, and the contrast is smaller. At the same time, the brightness component V in the HSV space is checked to confirm whether it meets the black edge characteristics (V mean < 15). Through structure verification, the real black edge and false alarm area can be effectively distinguished, and the accuracy of black edge detection is improved.

[0093] To improve the detection accuracy, multi-scale strip analysis is implemented. First, a wider strip (such as 64 pixels wide) is used for preliminary screening to locate the suspected area, and then a fine-grained strip (such as 8 pixels wide) is used within the suspected range for accurate positioning. The detection results of different scales are fused through a voting mechanism, and the final black border area needs to meet the consistency of at least two scales.

[0094] S5, geometric property constraint is performed on the accurately positioned area, the black border area is divided and labeled, and a structured report is generated.

[0095] Geometric constraints are added according to the characteristics of remote sensing images, including:

[0096] Morphological closing operation, morphological closing operation is performed on the accurately positioned black border area, discrete detection points are connected, and isolated noise areas are filtered. The closing operation can fill small holes in the area and connect adjacent areas through the operation of first expansion and then corrosion.

[0097] Non-rectangular image processing, for non-rectangular images (such as images after rotation and cropping), first perform minimum bounding rectangle calibration to convert the image to a rectangular area, then perform strip division and subsequent processing. The minimum bounding rectangle can tightly wrap the non-rectangular image to ensure the accuracy of strip division.

[0098] Generate a structured report, label the black border area coordinate range (represented by polygon vertices), and calculate the black border area ratio (BAD = black border pixel number / total pixel number x 100%).

[0099] Output quality grading results: Grade I (BAD < 1%), Grade II (1% ≤ BAD < 5%), and Grade III (BAD ≥ 5%). Additional visualization result layers are marked with semi-transparent red masks to mark the black border area, which facilitates manual review.

[0100] Then, according to the mutation judgment result, the specific position of the black border area is further judged in combination with the image edge structure.

[0101] The present application also provides a black border detection and removal device for remote sensing images, as shown in Figure 4 The device includes an image input module, a preprocessing module, a feature extraction module, a black border detection module, a geometric constraint and labeling module, and a report generation module, etc., wherein:

[0102] The image input module is used to receive remote sensing image data to be processed;

[0103] The preprocessing module performs preprocessing on the received remote sensing image, including denoising, contrast enhancement, etc., to improve the accuracy of subsequent processing;

[0104] A feature extraction module is configured to extract feature information related to black borders from the preprocessed remote sensing image, such as color, texture, etc.

[0105] A black border detection module is configured to detect, automatically identify and locate the black border region in the image based on the extracted feature information.

[0106] A geometric constraint and labeling module is configured to perform geometric property constraint on the detected black border region, divide the black border region and perform labeling.

[0107] A report generation module is configured to automatically generate a structured report including the coordinate range, area proportion and quality grading result of the black border region according to the processing result.

[0108] A visualization module is configured to display the processing result in a graphical manner, such as marking the black border region with a semi-transparent red mask, so as to facilitate intuitive understanding and review by the user.

[0109] By using the above device, an automatic, efficient and accurate remote sensing image black border detection and removal process can be realized. The device integrates multiple functional modules including image input, preprocessing, feature extraction, black border detection, geometric constraint and labeling, report generation and visualization, forming a complete processing chain. The user only needs to input the remote sensing image data to be processed into the image input module, and the device can automatically complete all subsequent processing steps without human intervention, greatly improving the processing efficiency. At the same time, since advanced machine learning or deep learning algorithms are used for black border detection and accurate geometric property constraint methods are used, the device can ensure the accuracy of black border detection and removal, and provide high-quality processing results for the user.

[0110] The application also provides an electronic device, Figure 5 The structure diagram of the electronic device provided by the embodiment of the application is shown in Figure 5 As shown in the figure, the electronic device can include a processor, a communications interface, a memory and a communications bus, wherein the processor, the communications interface and the memory complete mutual communication through the communications bus. The processor can call the logical instructions in the memory, for example, to execute the following method:

[0111] S1, strip division preprocessing is performed on the remote sensing image;

[0112] S2, multi-index parallel strategy is used to count the feature statistical value of each strip;

[0113] S3, a bidirectional mutation detection algorithm is used to compare the feature statistical values of adjacent strips to obtain a black border suspected region.

[0114] S4, after structure verification on the black edge suspected area, multi-scale strip fusion analysis is carried out to locate the black edge area;

[0115] S5, the geometric characteristic constraint is carried out to the accurate positioning area, the black edge area is divided and is labeled, and a structured report is generated.

[0116] In addition, the logic instructions in the memory described above can be realized in the form of a software functional unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part of the prior art that contributes or the part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in various embodiments of the present application. The storage medium described above includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0117] The embodiments of the present application also provide a non-transitory computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method provided by the above-mentioned embodiments, for example, including:

[0118] S1, strip division preprocessing is carried out on remote sensing images;

[0119] S2, multi-index parallel strategy is used to count each strip feature statistical value;

[0120] S3, adjacent strip feature statistical values are compared by using a bidirectional mutation detection algorithm to obtain a black edge suspected area;

[0121] S4, after structure verification on the black edge suspected area, multi-scale strip fusion analysis is carried out to locate the black edge area;

[0122] S5, the geometric characteristic constraint is carried out to the accurate positioning area, the black edge area is divided and is labeled, and a structured report is generated.

[0123] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0124] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0125] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for black border detection and removal for remote sensing imagery, characterized in that, The method comprises the following steps: S1, strip division preprocessing is performed on the remote sensing image; S2, multi-index parallel strategy is used to count the characteristic statistical value of each strip; S3, a bidirectional mutation detection algorithm is used to compare the characteristic statistical value of adjacent strips to obtain a black edge suspected area; S4, after structure verification of the black edge suspected area, multi-scale strip fusion analysis is performed to locate the black edge area; S5, geometric property constraint is performed on the accurately located area, the black edge area is divided and labeled, and a structured report is generated.

2. The method for black border detection and removal for remote sensing imagery according to claim 1, wherein, The remote sensing image category includes optical image, SAR image and infrared image, wherein the optical image is divided into RGB image, multispectral image or hyperspectral image; The strip division width is automatically adjusted according to the resolution of the remote sensing image, and the formula is: ； wherein, is the strip width, is the remote sensing image height, is the rounding function.

3. The method for black border detection and removal for remote sensing imagery according to claim 1, wherein, In the S2, the multi-index parallel strategy, the multi-index includes the mean (μ) and the standard deviation (σ) of the RGB or grayscale channel; In the S2, the multi-index parallel strategy, the multi-index includes the mean (μ) and the standard deviation (σ) of the RGB or grayscale channel; 4. The method for black border detection and removal for remote sensing imagery according to claim 1, wherein, In the S3, the bidirectional mutation detection algorithm is used, and when mutation detection is performed, an adaptive method based on global statistics is used to dynamically determine the comparison threshold, including the following steps: S31, the overall distribution of the multi-index characteristic value of all strips is calculated; S32, the median of the difference between the characteristic values of adjacent strips is taken as the reference threshold T; S33, the reference threshold is multiplied by an adjustment coefficient α according to the type of the remote sensing image, and a continuous mutation number threshold N is set; In the S3, the bidirectional mutation detection algorithm, the formula is: or ; wherein, is the mean μ threshold value, is the standard deviation σ threshold value; the black edge region warning is triggered when the mutation is detected for N consecutive times in the mutation region.

5. The method for black border detection and removal for remote sensing imagery according to claim 1, wherein, In the S4, the structure verification of the black edge suspected area includes: The Sobel edge feature map of the image is extracted, and the structure verification is performed according to the edge density, texture complexity and brightness component V of the HSV color space in the black edge warning area.

6. The method for black border detection and removal for remote sensing imagery of claim 1, wherein, In the S4, the multi-scale strip fusion analysis for locating the black edge area includes: S41, a wide-scale strip is used for preliminary screening and positioning of the suspected area; S42, a fine-scale strip is used for accurate positioning within the suspected range; S43, the detection results of different scales are fused through a voting mechanism to determine the black edge area.

7. The method for black border detection and removal for remote sensing imagery according to claim 1, wherein, In the S5, the geometric property constraint for the accurately located area includes: Discrete detection points are connected through morphological closing operation to filter isolated noise areas; For a non-rectangular image, strip division is performed after minimum circumscribed rectangle calibration.

8. The method for black border detection and removal for remote sensing imagery of claim 1, wherein, In the S5, the structured report includes: The black edge area ratio BAD is calculated, and the formula is: ； The quality classification result is output according to the black edge area ratio BAD, including: I level is BAD<1%, II level is 1%≤BAD<5%, and III level is (BAD≥5%), and a visual result layer is added, and the black edge area is marked with a semi-transparent red mask.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the steps of the black edge detection and removal method for remote sensing image according to any one of claims 1-8.

10. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the black edge detection and removal method for remote sensing image according to any one of claims 1-8.

Images