Remote sensing image cutting method and system based on multiband remote sensing image

Abstract

The invention relates to the technical field of image processing, and relates to a remote sensing image cutting method and system based on a multiband remote sensing image, and the method comprises the steps: carrying out the local superpixel region segmentation of an initial remote sensing image; performing region identification on the plurality of local superpixel regions to obtain at least one region of interest; according to the edge curve corresponding to each local superpixel region, performing region analysis on the at least one region of interest to obtain an edge detail disturbance coefficient corresponding to each region of interest; acquiring an edge diffusion coefficient corresponding to each region of interest; determining a cutting scale corresponding to each region of interest according to the edge detail disturbance coefficient corresponding to each region of interest and the edge diffusion coefficient corresponding to each region of interest; cutting each region of interest according to the cutting scale corresponding to each region of interest to obtain each cut region of interest; and splicing each cut region of interest to obtain a target remote sensing image. According to the invention, a high-quality target remote sensing image can be obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a remote sensing image cropping method and system based on multi-band remote sensing images. Background Art

[0002] With the rapid development of remote sensing technology, the amount and resolution of data that can be obtained have increased significantly. These high-resolution remote sensing data provide unprecedented rich information for a variety of application fields, such as geographic information systems, environmental monitoring, and urban planning. However, the massive growth of these data also brings challenges in data processing, especially in the cropping process of multi-band remote sensing images.

[0003] Cropping multi-band remote sensing images is a key step in remote sensing image processing, which involves extracting image blocks of specific areas or sizes from a large amount of raw remote sensing data. This step is crucial to improving the efficiency of data processing and ensuring the accuracy of analysis results. However, existing cropping methods mostly use a fixed-size window to move sequentially or randomly on images of different bands to crop small blocks of images for subsequent image reconstruction. This method ignores the significant differences in different band images in capturing ground information and detailed features, resulting in the inability to adapt to changes in local details during the cropping process, causing some key edge features to be lost or blurred after cropping, seriously affecting the quality and accuracy of the final image reconstruction. Summary of the invention

[0004] In order to solve the technical problem of detail information loss caused by differences between bands when cropping multi-band images, the purpose of the present invention is to provide a remote sensing image cropping method and system based on multi-band remote sensing images. The technical solution adopted is as follows: In a first aspect, an embodiment of the present invention provides a remote sensing image cropping method based on a multi-band remote sensing image, which is applied to a remote sensing image cropping system based on a multi-band remote sensing image. The method includes: Acquire initial remote sensing images of different bands, perform region segmentation processing on the initial remote sensing images, and obtain multiple local super-pixel regions with similar pixel features; Performing region recognition on the multiple local superpixel regions to obtain at least one region of interest; Performing edge detection on the multiple local super-pixel regions to obtain an edge curve corresponding to each local super-pixel region; Performing regional analysis on the at least one region of interest according to the edge curve corresponding to each local superpixel region to obtain an edge detail nuisance coefficient corresponding to each region of interest; Obtaining an edge diffusion coefficient corresponding to each region of interest, where the edge diffusion coefficient is an expansion coefficient in a same local superpixel region corresponding to each region of interest in different bands; Determining a cropping scale corresponding to each region of interest according to an edge detail nuisance coefficient corresponding to each region of interest and an edge diffusion coefficient corresponding to each region of interest; Cropping each region of interest according to the cropping scale corresponding to each region of interest to obtain each cropped region of interest; Each cropped region of interest is spliced ​​to obtain a target remote sensing image.

[0005] In a second aspect, an embodiment of the present invention provides a remote sensing image cropping system based on multi-band remote sensing images, comprising a memory and a processor, wherein the memory is connected to the processor, and the processor is used to execute one or more computer programs stored in the memory. When the processor executes the one or more computer programs, the remote sensing image cropping system based on multi-band remote sensing images implements the remote sensing image cropping method based on multi-band remote sensing images as described in the first aspect.

[0006] In a third aspect, an embodiment of the present invention provides a remote sensing image cropping device based on a multi-band remote sensing image, which is applied to a remote sensing image cropping system based on a multi-band remote sensing image, and the device includes: A processing unit, used for acquiring initial remote sensing images of different bands, performing region segmentation processing on the initial remote sensing images, and obtaining a plurality of local super-pixel regions with similar pixel features; an identification unit, configured to perform region identification on the plurality of local superpixel regions to obtain at least one region of interest; A detection unit, configured to perform edge detection on the plurality of local super-pixel regions to obtain an edge curve corresponding to each local super-pixel region; An analyzing unit, configured to perform a regional analysis on the at least one region of interest according to an edge curve corresponding to each local superpixel region, to obtain an edge detail nuisance coefficient corresponding to each region of interest; An acquisition unit, configured to acquire an edge diffusion coefficient corresponding to each region of interest, wherein the edge diffusion coefficient is an expansion coefficient in a same local superpixel region corresponding to each region of interest in different bands; A determining unit, configured to determine a cropping scale corresponding to each region of interest according to an edge detail nuisance coefficient corresponding to each region of interest and an edge diffusion coefficient corresponding to each region of interest; A cropping unit, configured to crop each region of interest according to a cropping scale corresponding to each region of interest, to obtain each cropped region of interest; The stitching unit is used to stitch each cropped region of interest to obtain a target remote sensing image.

[0007] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor executes the remote sensing image cropping method based on multi-band remote sensing images as described in the first aspect.

[0008] The present invention has the following beneficial effects: by acquiring initial remote sensing images of different bands and performing regional segmentation processing on them, the present invention can more accurately identify local super-pixel regions with similar pixel features, thereby improving the recognition accuracy of interest regions; further edge detection and regional analysis are performed on multiple local super-pixel regions, and the edge detail annoyance coefficient and edge diffusion coefficient of each interest region can be obtained. The edge detail annoyance coefficient and edge diffusion coefficient can help determine the cropping scale of each interest region, so that the cropping process can adapt to the characteristics of different interest regions, ensuring that the cropped region still retains important information, thereby retaining more edge details in the cropping process and reducing information loss; by splicing each interest region after cropping, the target remote sensing image can be effectively reconstructed, ensuring that the spliced ​​image has high integrity and consistency. Therefore, the method of the present invention can improve the accuracy and efficiency of remote sensing image processing while retaining more detail information, and finally obtain a higher quality target remote sensing image. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0010] Figure 1 A schematic flow chart of a remote sensing image cropping method based on multi-band remote sensing images provided by an embodiment of the present invention; Figure 2 A schematic structural diagram of a remote sensing image cropping device based on multi-band remote sensing images provided by one embodiment of the present invention; Figure 3 A schematic diagram of the structure of a remote sensing image cropping system based on multi-band remote sensing images provided by one embodiment of the present invention; Figure 4 A remote sensing image of one band provided by an embodiment of the present invention; Figure 5A remote sensing image of another band provided by an embodiment of the present invention; Figure 6 A corner point detection result diagram of a remote sensing image provided by an embodiment of the present invention; Figure 7 An edge detection result diagram of a remote sensing image provided by an embodiment of the present invention; Figure 8 for Figure 4 Image after adaptive cropping scale cropping; Fig. 9 for Figure 5 Image after adaptive cropping scale. DETAILED DESCRIPTION

[0011] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of a remote sensing image cropping method and system based on multi-band remote sensing images proposed by the present invention, its specific implementation method, structure, features and effects, in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0012] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0013] The specific scheme of a remote sensing image cropping method and system based on multi-band remote sensing images provided by the present invention is described in detail below in conjunction with the accompanying drawings.

[0014] See also Figure 1 , which shows a flow chart of a remote sensing image cropping method based on multi-band remote sensing images provided by an embodiment of the present invention, the method comprising the following steps: S10, acquiring initial remote sensing images of different bands, performing region segmentation processing on the initial remote sensing images, and obtaining a plurality of local super-pixel regions with similar pixel features.

[0015] Among them, the multi-band scanner carried by the drone and the ground sensor matrix can be used to obtain high-resolution multi-band remote sensing images of the target area. It is necessary to pay attention to the frequency and time interval of remote sensing image acquisition, such as obtaining data every 1 second, and the density of ground sensors, such as one sensor every 10 meters. Figure 4 and Figure 5 As shown in the figure, they are remote sensing images of different bands. Figure 4This is a remote sensing image selected in the short-wave infrared band (1.5-2.5 microns). The foreground information of this image is more prominent, the contour shadow is better, and the details are insufficient; Figure 5 This is a remote sensing image selected from the visible light to near-infrared band (0.7-1.1 microns). The image contains more detailed information, but the information is relatively messy.

[0016] Remote sensing images usually include multiple different bands, each of which captures reflection information in a different spectral range. Images in different bands can provide different types of ground information, for example, the visible light band can provide color information of the ground surface, and the infrared band can provide information on vegetation and water bodies.

[0017] Among them, the remote sensing images acquired in different bands are preprocessed, including radiation correction, geometric correction and atmospheric correction, so as to eliminate or reduce noise and errors in the images and ensure the accuracy of subsequent processing.

[0018] Among them, the region segmentation process is used to divide the image into several regions with similar characteristics. In remote sensing image processing, region segmentation can help identify and extract regions of objects with similar spectral characteristics. For example, the region segmentation algorithm can be a superpixel generation algorithm (such as SLIC, SEEDS, watersheds, etc.) to perform region segmentation on remote sensing images.

[0019] Among them, after the initial segmentation is completed, the segmentation results may need to be optimized by adjusting algorithm parameters or using post-processing techniques (such as merging small areas, splitting large areas, etc.) to improve the segmentation quality to ensure that the similarity of pixel features within each superpixel area is maximized, while the feature differences between regions are maximized.

[0020] Among them, the local superpixel region is a larger area composed of a series of pixels with similar features. Compared with a single pixel, a superpixel can provide a larger analysis unit, reduce the amount of data, and retain important spatial structure information. The aggregation of similar pixel features helps to identify the ground object category or target area with specific significance in the subsequent steps.

[0021] It can be seen that, in this embodiment, through this segmentation processing, the analysis of remote sensing images becomes more efficient and targeted, providing a basis for subsequent applications such as target recognition and feature extraction.

[0022] S20. Perform region recognition on the multiple local superpixel regions to obtain at least one region of interest.

[0023] Among them, region recognition refers to further identifying regions with specific characteristics or meeting specific conditions in the segmented superpixel regions, which may include identifying specific types of objects (such as water bodies, vegetation, buildings, etc.) or areas of interest. For example, the region recognition algorithm can be a supervised classification, an unsupervised classification, a rule-based classification, or a machine learning algorithm to identify superpixel regions and screen out regions that meet the definition criteria of the region of interest.

[0024] Among them, the region of interest refers to the area with specific significance or value determined in the process of region identification. In remote sensing image processing, the region of interest is usually the key area of ​​concern to researchers or users, such as the distribution area of ​​a certain type of land feature.

[0025] The interest level coefficient is a numerical value used to quantify the importance or degree of attention of each region of interest. The interest level coefficient can be determined based on a variety of factors, such as the spectral characteristics of the pixels in the region, the size of the region, the location of the region, etc. The interest level coefficient can be a numerical value between 0 and 1, or other forms of quantitative indicators, indicating the relative importance or interest level of the region as a region of interest.

[0026] Furthermore, the interest regions are sorted according to the interest coefficients, and those with higher coefficients are prioritized to improve work efficiency. When resources are limited, the interest coefficients can be used to determine which regions require more attention and resource investment. Alternatively, when analyzing the results, the interest coefficients can be used as a reference to help explain why some regions are more important or representative than others.

[0027] It can be seen that in this embodiment, regions with important features can be effectively identified from remote sensing images, and an interest level coefficient is assigned to each region, thereby providing important information and guidance for subsequent image analysis and processing.

[0028] S30, performing edge detection on the multiple local super-pixel regions to obtain an edge curve corresponding to each local super-pixel region.

[0029] Among them, edge detection aims to find the boundaries or contours of objects in the image, which usually indicate changes in the shape or area of ​​the object. In remote sensing images, edge detection can be used to identify the boundaries of different types of objects, such as the boundary between water and land, the boundary between vegetation and non-vegetation areas, etc. Optionally, the edge detection algorithm can be a Canny edge detection operator, which identifies edges by calculating the gradient amplitude and direction of the image, or an edge detection algorithm based on deep learning, such as HED (Holistically-nested Edge Detection) or RCF (Richer Convolutional Features), which can provide better detection effects when processing complex scenes.

[0030] Optionally, before edge detection, it is usually necessary to preprocess multiple local superpixel regions, including grayscale conversion, noise removal, etc., to improve the accuracy of edge detection.

[0031] In the specific implementation of edge detection, the gradient of each pixel point in each local superpixel area in multiple local superpixel areas is calculated. The magnitude of the gradient represents the rate of change of brightness, and the direction of the gradient points to the direction where the brightness increases fastest. Commonly used gradient operators include the Sobel operator, the Prewitt operator, and the Canny operator. After the gradient is calculated, non-maximum suppression is performed to ensure that only local maximum points in the gradient direction are retained, so that the edge can be refined. By setting high and low thresholds, edge points can be connected to form a complete edge curve. Figure 7 As shown, the figure shows the edge detection result of the remote sensing image.

[0032] Therefore, the edge curve is composed of a series of edge points, which depict the outer contour of the area. The edge can be a step edge, where the brightness suddenly jumps from one value to another, or a roof edge, where the brightness increases first and then decreases, or decreases first and then increases.

[0033] Among them, different edge detection methods or parameters may need to be used for features of different scales. For example, filters of different sizes may be used to detect coarse edges and subtle edges.

[0034] It can be seen that in this embodiment, edge curves are effectively extracted from each local superpixel region, providing important structural information for subsequent image analysis and processing.

[0035] S40. Performing a regional analysis on the at least one region of interest according to the edge curve corresponding to each local superpixel region to obtain an edge detail nuisance coefficient corresponding to each region of interest.

[0036] The purpose of regional analysis is to gain a deeper understanding of the characteristics of the region of interest, including the details of its edges, shape, texture, etc. In remote sensing applications, it is used to identify specific types of objects or monitor changes in objects.

[0037] Among them, the edge detail annoyance coefficient is an indicator to measure the edge clarity and integrity. It can be calculated by analyzing the continuity and integrity of the edge curve and the distribution of edge pixels. The edge detail annoyance coefficient reflects the significance and reliability of edge features.

[0038] It can be seen that in this embodiment, by calculating the edge detail nuisance coefficient, the characteristics of the region of interest in the remote sensing image can be better understood, and a quantitative basis can be provided for subsequent processing and analysis.

[0039] S50, obtaining an edge diffusion coefficient corresponding to each region of interest, where the edge diffusion coefficient is an expansion coefficient in a same local superpixel region corresponding to each region of interest in different bands.

[0040] The edge diffusion coefficient quantifies the diffusion degree of the edge of the region of interest in different bands. The edge diffusion coefficient is used to understand the spatial distribution characteristics of the region of interest in different bands, and their expansion or contraction with the change of bands.

[0041] For example, if the edge diffusion coefficient is large, this may indicate that the region of interest has significant changes in shape or size across different bands.

[0042] In one embodiment, obtaining the edge diffusion coefficient corresponding to each region of interest includes: obtaining a third region of interest, a third remote sensing image of a first band where the third region of interest is located, and a fourth remote sensing image of a second band where the third region of interest is located, wherein the third region of interest is any region of interest among each region of interest, and the first band and the second band are adjacent bands; in the third remote sensing image, calculating the coordinates of all first edge curve pixel points corresponding to the third region of interest and the first centroid coordinates to obtain a first distance; in the fourth remote sensing image, calculating the coordinates of all second edge curve pixel points corresponding to the third region of interest and the second centroid coordinates to obtain a second distance; calculating the first distance and the second distance to obtain the edge diffusion coefficient corresponding to the third region of interest; and calculating each region of interest in turn to obtain the edge diffusion coefficient corresponding to each region of interest.

[0043] Among them, a third region of interest (third region of interest) is selected from the identified regions of interest for analysis, and a remote sensing image (third remote sensing image) of the third region of interest in the first band and a remote sensing image (fourth remote sensing image) in the second band are obtained. The first band and the second band are adjacent bands, and are usually closely arranged in the spectrum.

[0044] Specifically, in the third remote sensing image, the distance between the coordinates of all first edge curve pixel points corresponding to the third region of interest and the coordinates of the centroid of the third region of interest (first centroid coordinates) is calculated to obtain the first distance; in the fourth remote sensing image, the distance between the coordinates of all second edge curve pixel points corresponding to the third region of interest and the coordinates of the centroid of the third region of interest (second centroid coordinates) is calculated to obtain the second distance. The first distance and the second distance can be calculated using a variety of methods, such as Euclidean distance, Manhattan distance, etc.

[0045] The first distance and the second distance may be calculated according to the following formula: ; In the formula, The cth band image represents the third area of ​​interest. Represents The dth band image with adjacent bands, Represents the cth band image The Euclidean distance between the coordinates of all edge curve pixels and the centroid coordinates in the third region of interest, Represents the dth band image The Euclidean distance between the coordinates of all edge curve pixels and the centroid coordinates in the third region of interest, Represents the The edge diffusion coefficient of the third region of interest.

[0046] It should be noted that due to It is calculated by the difference between two distances, so it is possible to have a negative number. For example, if in the c-th band image, Less than hour, is a negative number. Therefore, when A positive number indicates that in the c-th band image, the distance between the edge curve pixel point and the centroid coordinate is large, indicating that the edge diffusion in this band is large; A negative number means that in the dth band image, the distance between the edge curve pixel point and the centroid coordinate is large, indicating that the edge diffusion under this band is large.

[0047] It can be seen that in this embodiment, by calculating the edge diffusion coefficient, the characteristics of the region of interest in the remote sensing image in different bands can be better understood, and a basis can be provided for further analysis and processing of the image.

[0048] S60: Determine a cropping scale corresponding to each region of interest according to an edge detail nuisance coefficient corresponding to each region of interest and an edge diffusion coefficient corresponding to each region of interest.

[0049] The cropping scale refers to the degree of detail and size that should be retained in each region of interest when cropping remote sensing images. This scale is crucial to retaining important information in the image while removing redundant or noisy information.

[0050] In one embodiment, determining the cropping scale corresponding to each region of interest according to the edge detail disturbance coefficient corresponding to each region of interest and the edge diffusion coefficient corresponding to each region of interest includes: obtaining the edge detail loss rate corresponding to each region of interest according to the edge detail disturbance coefficient corresponding to each region of interest and the edge diffusion coefficient corresponding to each region of interest; obtaining the detail feature loss factor corresponding to each region of interest according to the edge detail loss rate corresponding to each region of interest and the corresponding interest degree coefficient carried by each region of interest; and determining the cropping scale corresponding to each region of interest according to the detail feature loss factor corresponding to each region of interest.

[0051] Wherein, in obtaining the edge detail loss rate corresponding to each region of interest according to the edge detail annoyance coefficient corresponding to each region of interest and the edge diffusion coefficient corresponding to each region of interest, it can be calculated according to the following formula: ; In the formula, Represents the The edge detail annoyance coefficient of the region of interest, Represents the The edge diffusion coefficient of the region of interest is Represents the The edge detail loss rate of the region of interest.

[0052] Among them, the edge detail loss rate is an indicator of the variation and disturbance degree of the edge details of the interest area between bands.

[0053] Among them, in obtaining the detail feature loss factor corresponding to each region of interest according to the edge detail loss rate corresponding to each region of interest and the interest degree coefficient corresponding to each region of interest, it can be calculated according to the following formula: ; In the formula, Represents the The regional interest level coefficient of the interest area, Represents the The edge detail loss rate of the region of interest, is the normalization function, Represents the The detail feature loss factor of the interest region.

[0054] in, It is a normalization function, which is used to standardize the results to a certain range. Represents the The detail feature loss factor of the interest region represents the edge detail loss and the importance of the region.

[0055] In the process of determining the cropping scale corresponding to each region of interest according to the detail feature loss factor corresponding to each region of interest, the pre-cropping window size scale of the target region of interest is obtained. , and the detail feature loss factor of the target interest area are further calculated according to the following formula: ; In the formula, Represents the The detail feature loss factor of the region of interest, Represents the The pre-cropped window size scale of the region of interest, Represents the The cropping scale of the region of interest.

[0056] Among them, when When it is not an integer, Perform numerical processing on, such as rounding: Round to the nearest integer value; or take the integer directly The integer part of , ignoring the decimal part; or interpolation method, if more precise cropping is required, the interpolation method can be used to crop and scale the image to adapt to the non-integer cropping scale. The above numerical processing method is not limited here.

[0057] Among them, the cutting scale Represents a scaling factor, which represents the scale relative to the pre-clipped window size ratio, and therefore scales with the pre-clipped window size Multiply the length and width to get the final crop size.

[0058] Specifically, pre-cut the window size scale is a rectangular window with a length of A1 and a width of B1. The final crop size can be calculated as follows: , new width ; Therefore, the cropped window size is .

[0059] Among them, the pre-set cutting window size scale It can be set according to the application purpose and requirements of the remote sensing image. It is a reference value used in combination with the detail feature loss factor to determine the final cropping scale.

[0060] The cropping scale is inversely proportional to the detail feature loss factor of the region of interest. If the detail feature loss factor is high, it means that the region may lose more details during the cropping process, so the cropping scale will be reduced accordingly to retain more information.

[0061] It can be seen that in this embodiment, a suitable cropping scale for each region of interest is calculated to ensure that important edge and detail features are retained as much as possible during the cropping process, while improving the efficiency and accuracy of data processing.

[0062] S70: Crop each region of interest according to the cropping scale corresponding to each region of interest to obtain each cropped region of interest.

[0063] Among them, for each region of interest, the corresponding cropping scale is used to determine the size and shape of the cropping region. The cropping scale may be adjusted according to the characteristics of the region of interest. For example, if the detail feature loss factor of a certain area is high, the cropping scale may need to be reduced to retain more details.

[0064] The cropping operation may be a simple pixel extraction or may involve more complex image processing techniques, such as maintaining the resolution of the image, smoothing or sharpening the image, etc.

[0065] Specifically, use image processing software or custom algorithms to crop each region of interest according to the cropping scale. The specific operations involved are as follows: determine the size and position of the cropping window according to the cropping scale; extract the sub-image covered by the cropping window from the original remote sensing image, and this sub-image is the cropped region of interest; save each cropped region of interest as an independent image file or image data set for subsequent analysis and processing. Figure 8 and Fig. 9 As shown, Figure 8 for Figure 4 The image after adaptive cropping scale cropping, Figure 8 The edges of the cropping window are smoother and can better represent the local areas of different features in the remote sensing image of the target open-pit mine. Fig. 9 for Figure 5 The image after adaptive cropping scale cropping, Fig. 9 The edge of the cropping window appears jagged, which can better show the detailed features inside the local feature area.

[0066] It can be seen that in this embodiment, it is ensured that each cropped region of interest contains the most important information, while unnecessary background or other interfering information is removed, thereby improving the accuracy and efficiency of remote sensing image analysis.

[0067] S80, stitching each cropped region of interest to obtain a target remote sensing image.

[0068] After cropping each region of interest, these cropped regions are reassembled to form a new, more compact remote sensing image, namely, the target remote sensing image.

[0069] Among them, it is necessary to ensure that all cropped regions of interest have the same coordinate system.

[0070] Specifically, the cropped regions of interest are stitched according to their relative positions in the original images, including the following steps: ensuring that all regions of interest are aligned when stitching, which usually involves geometric correction and registration; merging the regions of interest into a single image plane to form a complete remote sensing image, and creating a mask to process the overlapping parts between the regions of interest to ensure that there is no duplicate or missing data in the stitched image; checking the stitched image to ensure that there are no obvious seams or errors.

[0071] Optionally, save the stitched target remote sensing image as a new image file or dataset, ensuring that the saved image contains all necessary information, such as metadata, coordinate system, etc., for subsequent use and analysis.

[0072] It can be seen that in this embodiment, the integrity of the remote sensing image is restored through stitching, and the stitched image reduces the problem of edge information loss that may occur during the cropping and segmentation process.

[0073] The present invention has the following beneficial effects: by acquiring initial remote sensing images of different bands and performing regional segmentation processing on them, the present invention can more accurately identify local super-pixel regions with similar pixel features, thereby improving the recognition accuracy of interest regions; further edge detection and regional analysis are performed on multiple local super-pixel regions, and the edge detail annoyance coefficient and edge diffusion coefficient of each interest region can be obtained. The edge detail annoyance coefficient and edge diffusion coefficient can help determine the cropping scale of each interest region, so that the cropping process can adapt to the characteristics of different interest regions, ensuring that the cropped region still retains important information, thereby retaining more edge details in the cropping process and reducing information loss; by splicing each interest region after cropping, the target remote sensing image can be effectively reconstructed, ensuring that the spliced ​​image has high integrity and consistency. Therefore, the method of the present invention can improve the accuracy and efficiency of remote sensing image processing while retaining more detail information, and finally obtain a higher quality target remote sensing image.

[0074] In one embodiment, the region identification is performed on the multiple local superpixel regions to obtain at least one region of interest, including: performing corner point detection processing on the multiple local superpixel regions to obtain all feature corner points in the multiple local superpixel regions; performing calculations based on all feature corner points in the multiple local superpixel regions to obtain an interest level coefficient corresponding to each local superpixel region; comparing the interest level coefficient corresponding to each local superpixel region with a preset interest level threshold, screening out local superpixel regions greater than the preset interest level threshold as regions of interest, and obtaining at least one region of interest.

[0075] The above-mentioned corner detection process is used to identify points with significant features in the image, which have high contrast changes in the geometric structure of the image. The corner detection process can use the Shi-Tomasi corner detection algorithm, which is based on the Taylor series expansion of the image grayscale and determines the corners by calculating the minimum eigenvalue of each point in the image. This is not a unique limitation. Figure 6 As shown, the figure shows the corner detection results of the remote sensing image.

[0076] The preset interest level threshold may be set manually or based on experience to filter out areas with a higher interest level. For example, it may be set to 0.5, which is not a unique limitation.

[0077] It can be seen that in this embodiment, regions with important features can be effectively identified from remote sensing images, and an interest level coefficient is assigned to each region, thereby providing important information and guidance for subsequent image analysis and processing.

[0078] Specifically, the calculation is performed based on all feature corner points in the multiple local superpixel areas to obtain the interest degree coefficient corresponding to each local superpixel area, including: obtaining a target local superpixel area and a first corner point set corresponding to the target local superpixel area, the target local superpixel area is any area in the multiple local superpixel areas, and the first corner point set includes all corner points in the target local superpixel area; obtaining a first remote sensing image of the target local superpixel area at a first moment and a second remote sensing image at a second moment, the second moment being an adjacent left moment to the first moment; obtaining a first target local superpixel corner point set corresponding to the first remote sensing image and a first target local superpixel corner point set corresponding to the second remote sensing image a second target local superpixel corner point set; obtaining the corner point difference of the target local superpixel area in the first remote sensing image and the second remote sensing image according to the first target local superpixel corner point set and the second target local superpixel corner point set; performing distance calculation according to the first corner point set corresponding to the target local superpixel area to obtain the mean Euclidean distance between all corner points in the target local superpixel area; calculating the mean Euclidean distance between all corner points in the target local superpixel area and the corner point difference to obtain the interest degree coefficient corresponding to the target local superpixel area; performing calculation on each local superpixel area in turn to obtain the interest degree coefficient corresponding to each local superpixel area.

[0079] Among them, in calculating the mean Euclidean distance between all corner points in the target local superpixel area and the corner point difference, the interest degree coefficient corresponding to the target local superpixel area can be calculated according to a first preset formula, and the first preset formula is as follows: ; In the formula, Represents the first local superpixel region, represents the remote sensing image at time a in the time series, and b represents the remote sensing image at the left moment adjacent to time a in the time series. Represents the The difference between the corner points of the local superpixel region in the image at adjacent moments a and b, Represents the The mean Euclidean distance between all corner points in the local superpixel region, Represents the The interest degree coefficient of the superpixel region, e represents a natural constant.

[0080] Among them, one of the multiple local superpixel regions obtained by the superpixel segmentation algorithm is selected as the target local superpixel region. The target local superpixel region is a sub-region after image segmentation and has similar pixel features.

[0081] Furthermore, for the selected target local superpixel region, a corner detection algorithm (such as the Shi-Tomasi algorithm) is used to identify and obtain all corner points in the region to form a first corner point set.

[0082] The first moment and the second moment are adjacent time points, and the second moment is the adjacent left moment of the first moment.

[0083] As shown in the first preset formula, the first remote sensing image at the first moment is expressed in the first preset formula as , the second remote sensing image at the second moment is expressed as b in the first preset formula.

[0084] The corner point difference reflects the change of the target local superpixel area between two moments, and is regarded as the structural steady-state coefficient of the target local superpixel area. The larger the value, the more dynamic the topographic features of the target local superpixel area. In specific implementation: the corner point difference can be achieved by matching the corner points in the first and second corner point sets, which involves a corner point matching algorithm, such as FLANN or Brute-Force matcher, to determine the corresponding corner points in the two sets and calculate their position changes.

[0085] Among them, the mean Euclidean distance between all corner points in the target local superpixel area reflects the compactness of the corner point distribution in the area. The smaller the value, the more significant the geomorphic features of the area.

[0086] The Euclidean distance is the straight-line distance between two points in Euclidean space. For a set of corner points, it means calculating the distance between each pair of corner points in the set and then finding the average of all distances.

[0087] Specifically, the above steps are repeated for all local superpixel regions, and the corresponding interest degree coefficient is calculated for each region.

[0088] It can be seen that in this embodiment, an interest degree coefficient can be assigned to each local superpixel region, which quantifies the feature change and compactness of the region, thereby helping to identify regions with significant feature changes in the image. The mean Euclidean distance between all corner points in the target local superpixel region and the corner point difference are calculated.

[0089] In one embodiment, the performing of regional analysis on the at least one region of interest according to the edge curve corresponding to each local superpixel region to obtain the edge detail disturbance coefficient corresponding to each region of interest includes: performing feature analysis on the at least one region of interest according to the edge curve corresponding to each local superpixel region to obtain the feature value corresponding to each region of interest; determining the target cropping band corresponding to each region of interest in different bands according to the feature value corresponding to each region of interest; and obtaining the edge detail disturbance coefficient corresponding to each region of interest according to the target cropping band corresponding to each region of interest.

[0090] Among them, feature analysis aims to extract numerical values ​​or parameters that describe regional characteristics to obtain feature values, which may include edge gradient amplitude, edge direction, edge continuity, etc. The feature value reflects the significance and reliability of the edge.

[0091] Among them, in multi-band remote sensing images, different bands may contain different ground object information. Through feature analysis, it is further determined which bands are most important for a specific area of ​​interest.

[0092] Specifically, according to the eigenvalues ​​corresponding to each region of interest, the band that best represents the characteristics of the region is selected as the target clipping band. For example, if an area of ​​interest is mainly water, then the band related to the reflectivity of the water body may be selected. Band selection can be based on a variety of criteria, such as the maximum discrimination of the band, the minimum noise, and the correlation with a specific type of ground object.

[0093] It can be seen that in this embodiment, a detailed edge analysis is performed on each local superpixel region to obtain the edge detail nuisance coefficient corresponding to each region of interest, which provides a solid foundation for further processing and analysis of remote sensing images.

[0094] In one embodiment, the feature analysis of the at least one region of interest is performed according to the edge curve corresponding to each local superpixel region to obtain the feature value corresponding to each region of interest, including: obtaining a target region of interest and a target edge curve in which the target region of interest is located, wherein the target region of interest is any region of the at least one region of interest; obtaining the total number of pixels in the target region of interest and the number of edge pixels on the target edge curve in which the target region of interest is located; obtaining an edge density coefficient of the target region of interest according to the total number of pixels in the target region of interest and the number of edge pixels on the target edge curve in which the target region of interest is located; obtaining all local superpixel regions adjacent to the target region of interest to obtain a first local superpixel region set; calculating the total number of pixels in the target region of interest to obtain a mean pixel value corresponding to the target region of interest; obtaining a degree of interest coefficient corresponding to each first local superpixel region in the first local superpixel region set; calculating the edge density coefficient of the target region of interest, the mean pixel value corresponding to the target region of interest, and the degree of interest coefficient corresponding to each first local superpixel region to obtain a feature value corresponding to the target region of interest; and performing calculations on each region of interest in turn to obtain a feature value corresponding to each region of interest.

[0095] Among them, in the process of obtaining the edge density coefficient of the target interest area according to the total number of pixels in the target interest area and the number of edge pixels on the target edge curve where the target interest area is located, the edge density coefficient of the target interest area = the number of edge pixels on the target edge curve where the target interest area is located / the total number of pixels in the target interest area.

[0096] Therefore, the edge density coefficient is a ratio, which represents the ratio of the number of edge pixels in the target interest region to the total number of pixels. The larger the edge density coefficient of the target interest region, the more obvious the edge features of the target region are and the richer the details are.

[0097] The mean pixel value reflects the average brightness or color of the target area of ​​interest.

[0098] The local superpixel regions in the first local superpixel region set share edges or features with the target region of interest.

[0099] Wherein, in the process of calculating the edge density coefficient of the target interest region, the mean pixel value corresponding to the target interest region and the interest degree coefficient corresponding to each first local superpixel region, the calculation can be performed according to the second preset formula, and the second preset formula is as follows: ; In the formula, Represents the first target area of ​​interest, Represents the The edge density coefficient of the target interest area, Represents the The set of all superpixel regions closely connected to the target interest region, that is, the first local superpixel region set, Represents The first in the collection local superpixel region, Represents the The mean pixel value of the target interest area, Represents the The average pixel value of a local superpixel area, Represents the The interest coefficient of the local superpixel region, Represents the The feature value of the target area of ​​interest.

[0100] The above steps are repeated to calculate each region of interest using the second preset formula to obtain the characteristic value corresponding to each region of interest.

[0101] For example, a multi-band remote sensing image containing red, green, blue and near-infrared bands is taken, and the SLIC (Simple Linear Iterative Clustering) algorithm is used to perform superpixel segmentation on the image to obtain 100 local superpixel regions; Shi-Tomasi corner point detection is performed on each local superpixel region to obtain the characteristic corner points in each region, and the interest degree coefficient of each local superpixel region is calculated based on the corner point difference and the mean Euclidean distance. The interest degree threshold is set to 0.5 and 10 interest regions are identified; Canny edge detection is performed on each interest region to obtain the edge curve of each region; a target interest region (for example, interest region 1) is selected for feature analysis; the target There are 1000 pixels in the interest region 1, of which the number of edge pixels is 200, and the edge density coefficient = 200 / 1000 = 0.2. Five local superpixel regions adjacent to the target interest region 1 are determined to form a first local superpixel region set; the mean pixel value calculation result of the target interest region 1 is 100 (assumed value), and the interest degree coefficients of each region in the current first local superpixel region set are 0.6, 0.55, 0.4, 0.65, and 0.3, respectively. According to the second preset formula, the eigenvalue of the target interest region 1 is 2.4124.

[0102] It can be seen that this embodiment can perform detailed feature analysis on each region of interest to obtain the feature value corresponding to each region, thereby providing a solid foundation for further processing and analysis of remote sensing images.

[0103] In one embodiment, determining the target cropping band corresponding to each region of interest in different bands based on the characteristic value corresponding to each region of interest includes: obtaining multiple first characteristic values ​​of a first region of interest in different bands, where the first region of interest is any region of interest among each region of interest; arranging the multiple first characteristic values ​​according to a preset arrangement rule to obtain a target sequence; selecting the first characteristic value with the largest value in the target sequence as the target characteristic value; determining the initial remote sensing image corresponding to the target characteristic value as the first target cropping band corresponding to the first region of interest; and performing band processing on each region of interest in turn to obtain the target cropping band corresponding to each region of interest.

[0104] The target sequence can be ;in, Represents the total number of Band remote sensing images.

[0105] Among them, the target eigenvalue represents the most significant characteristic performance of the area of ​​interest in all bands.

[0106] The preset arrangement rule may be to arrange in descending order according to the size of the characteristic values, and there is no unique restriction here.

[0107] Specifically, assuming that there are P bands of remote sensing images, feature value extraction is performed on the first area of ​​interest on P bands, and P feature values ​​are obtained, which are ,in is the eigenvalue of the first band, is the eigenvalue of the second band, and so on until is the eigenvalue of the Pth band. According to the descending order of the eigenvalue, the target sequence E will be; ,in, is the largest eigenvalue, is the smallest eigenvalue. In the target sequence E, the eigenvalue with the largest value is selected and this value will be used as the target eigenvalue. In the sequence E, since it is already arranged in descending order, is the target eigenvalue; target eigenvalue The corresponding initial remote sensing image band is the first target cropping band corresponding to the first area of ​​interest, indicating that the first area of ​​interest has the most significant features in the first band, so this band is selected for subsequent cropping processing.

[0108] Therefore, for each region of interest, the above process is repeated to determine the target feature value of each region of interest in different bands, and accordingly determine the target cropping band corresponding to each region of interest.

[0109] It can be seen that in this embodiment, all regions of interest determine their respective target clipping bands, thereby performing more accurate remote sensing image clipping and analysis.

[0110] In one embodiment, the edge detail disturbance coefficient corresponding to each region of interest is obtained according to the target cropping band corresponding to each region of interest, including: obtaining a second region of interest and a second target cropping band corresponding to the second region of interest, the second region of interest being any region of interest among each region of interest; obtaining all local superpixel regions adjacent to the second region of interest to obtain a second local superpixel region set; calculating the eigenvalue of the second local superpixel region set in the second target cropping band corresponding to the second region of interest to obtain a second eigenvalue; calculating the second eigenvalue and the second target cropping band corresponding to the second region of interest to obtain the edge detail disturbance coefficient of the second region of interest; and calculating each region of interest in turn to obtain the edge detail disturbance coefficient corresponding to each region of interest.

[0111] In the process of calculating the second eigenvalue and the second target clipping band corresponding to the second region of interest, the calculation may be performed according to a third preset formula, and the third preset formula is as follows: ; in, Represents the The optimal cropping band for the second region of interest, Represents the The local superpixel region is The characteristic values ​​in the band image, Represents the The edge detail nuisance coefficient of the second region of interest.

[0112] The second target cropping band refers to the most significant feature expression of the second region of interest in different bands determined in the preceding step.

[0113] The selection of the adjacent regions is usually based on spatial relationships, such as pixel adjacency or a specific distance threshold. The local superpixel regions in the second local superpixel region set share edges or features with the second region of interest.

[0114] The second eigenvalue represents the significance of each local superpixel region in the target cropped band.

[0115] The edge detail nuisance coefficient of the second region of interest is obtained by calculating the average value of the eigenvalues ​​of all local superpixel regions in the second local superpixel region set in the optimal cropping band.

[0116] It can be seen that in this embodiment, an edge detail annoyance coefficient is calculated for each region of interest, and the edge detail annoyance coefficient reflects the complexity and variation degree of the edge of the region of interest.

[0117] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0118] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0119] As another aspect of the embodiment of the present application, the embodiment of the present application provides a remote sensing image cropping device based on multi-band remote sensing images. The remote sensing image cropping device based on multi-band remote sensing images can be a software module, and the software module includes a number of instructions, which are stored in a memory, and the processor can access the memory and call the instructions for execution to complete the remote sensing image cropping method based on multi-band remote sensing images described in the above-mentioned various embodiments.

[0120] See also Figure 2 , Figure 2 Schematic diagram of a remote sensing image cropping device based on multi-band remote sensing images provided in an embodiment of the present application. Figure 2 As shown, the remote sensing image cropping device 200 based on multi-band remote sensing images includes: The processing unit 201 is used to obtain initial remote sensing images of different bands, and perform region segmentation processing on the initial remote sensing images to obtain multiple local super-pixel regions with similar pixel features; An identification unit 202, configured to perform region identification on the plurality of local superpixel regions to obtain at least one region of interest; A detection unit 203 is used to perform edge detection on the multiple local super-pixel regions to obtain an edge curve corresponding to each local super-pixel region; An analyzing unit 204 is configured to perform a regional analysis on the at least one region of interest according to an edge curve corresponding to each local superpixel region, to obtain an edge detail nuisance coefficient corresponding to each region of interest; An acquiring unit 205 is configured to acquire an edge diffusion coefficient corresponding to each region of interest, where the edge diffusion coefficient is an expansion coefficient in a same local superpixel region corresponding to each region of interest in different bands; A determination unit 206, configured to determine a cropping scale corresponding to each region of interest according to an edge detail nuisance coefficient corresponding to each region of interest and an edge diffusion coefficient corresponding to each region of interest; A cropping unit 207, configured to crop each region of interest according to a cropping scale corresponding to each region of interest to obtain each cropped region of interest; The stitching unit 208 is used to stitch each cropped region of interest to obtain a target remote sensing image.

[0121] The present invention has the following beneficial effects: by acquiring initial remote sensing images of different bands and performing regional segmentation processing on them, the present invention can more accurately identify local super-pixel regions with similar pixel features, thereby improving the recognition accuracy of interest regions; further edge detection and regional analysis are performed on multiple local super-pixel regions, and the edge detail annoyance coefficient and edge diffusion coefficient of each interest region can be obtained. The edge detail annoyance coefficient and edge diffusion coefficient can help determine the cropping scale of each interest region, so that the cropping process can adapt to the characteristics of different interest regions, ensuring that the cropped region still retains important information, thereby retaining more edge details in the cropping process and reducing information loss; by splicing each interest region after cropping, the target remote sensing image can be effectively reconstructed, ensuring that the spliced ​​image has high integrity and consistency. Therefore, the method of the present invention can improve the accuracy and efficiency of remote sensing image processing while retaining more detail information, and finally obtain a higher quality target remote sensing image.

[0122] In one embodiment, in performing region recognition on the multiple local superpixel regions to obtain at least one region of interest, the recognition unit 202 is further used to: perform corner point detection processing on the multiple local superpixel regions to obtain all feature corner points in the multiple local superpixel regions; perform calculations based on all feature corner points in the multiple local superpixel regions to obtain an interest level coefficient corresponding to each local superpixel region; compare the interest level coefficient corresponding to each local superpixel region with a preset interest level threshold, screen out local superpixel regions greater than the preset interest level threshold as regions of interest, and obtain at least one region of interest.

[0123] In one embodiment, in the step of calculating based on all feature corner points in the multiple local superpixel areas to obtain the interest level coefficient corresponding to each local superpixel area, the identification unit 202 is further used for: the step of calculating based on all feature corner points in the multiple local superpixel areas to obtain the interest level coefficient corresponding to each local superpixel area includes: obtaining a target local superpixel area and a first corner point set corresponding to the target local superpixel area, the target local superpixel area being any area in the multiple local superpixel areas, the first corner point set including all corner points in the target local superpixel area; obtaining a first remote sensing image of the target local superpixel area at a first moment and a second remote sensing image at a second moment, the second moment being a moment adjacent to the left of the first moment; obtaining the A first target local superpixel corner point set corresponding to the first remote sensing image and a second target local superpixel corner point set corresponding to the second remote sensing image; obtaining the corner point difference of the target local superpixel area in the first remote sensing image and the second remote sensing image according to the first target local superpixel corner point set and the second target local superpixel corner point set; performing distance calculation according to the first corner point set corresponding to the target local superpixel area to obtain the mean Euclidean distance between all corner points in the target local superpixel area; calculating the mean Euclidean distance between all corner points in the target local superpixel area and the corner point difference to obtain the interest degree coefficient corresponding to the target local superpixel area; performing calculation on each local superpixel area in turn to obtain the interest degree coefficient corresponding to each local superpixel area.

[0124] In one embodiment, in performing regional analysis on the at least one region of interest according to the edge curve corresponding to each local superpixel region to obtain the edge detail annoyance coefficient corresponding to each region of interest, the analysis unit 204 is further used to: perform feature analysis on the at least one region of interest according to the edge curve corresponding to each local superpixel region to obtain the feature value corresponding to each region of interest; determine the target cropping band corresponding to each region of interest in different bands according to the feature value corresponding to each region of interest; and obtain the edge detail annoyance coefficient corresponding to each region of interest according to the target cropping band corresponding to each region of interest.

[0125] In one embodiment, in the step of performing feature analysis on the at least one region of interest according to the edge curve corresponding to each local superpixel region to obtain a feature value corresponding to each region of interest, the analysis unit 204 is further used to: obtain a target region of interest and a target edge curve where the target region of interest is located, the target region of interest being any region of the at least one region of interest; obtain a total number of pixels in the target region of interest and a number of edge pixels on the target edge curve where the target region of interest is located; obtain the target region of interest according to the total number of pixels in the target region of interest and the number of edge pixels on the target edge curve where the target region of interest is located. The edge density coefficient of the target area of ​​interest is obtained; all local superpixel areas adjacent to the target area of ​​interest are obtained to obtain a first local superpixel area set; the total number of pixels in the target area of ​​interest is calculated to obtain the mean pixel value corresponding to the target area of ​​interest; the interest degree coefficient corresponding to each first local superpixel area in the first local superpixel area set is obtained; the edge density coefficient of the target area of ​​interest, the mean pixel value corresponding to the target area of ​​interest and the interest degree coefficient corresponding to each first local superpixel area are calculated to obtain the eigenvalue corresponding to the target area of ​​interest; each area of ​​interest is calculated in turn to obtain the eigenvalue corresponding to each area of ​​interest.

[0126] In one embodiment, in determining the target cropping band corresponding to each region of interest in different bands according to the eigenvalue corresponding to each region of interest, the analysis unit 204 is further used to: obtain multiple first eigenvalues ​​of the first region of interest in different bands, the first region of interest being any region of interest among each region of interest; arranging the multiple first eigenvalues ​​according to a preset arrangement rule to obtain a target sequence; screening out the first eigenvalue with the largest value in the target sequence as the target eigenvalue; determining the initial remote sensing image corresponding to the target eigenvalue as the first target cropping band corresponding to the first region of interest; and performing band processing on each region of interest in turn to obtain the target cropping band corresponding to each region of interest.

[0127] In one embodiment, in obtaining the edge detail nuisance coefficient corresponding to each region of interest according to the target cropping band corresponding to each region of interest, the analysis unit 204 is further used to: obtain a second region of interest and a second target cropping band corresponding to the second region of interest, where the second region of interest is any region of interest in each region of interest, obtain all local superpixel regions adjacent to the second region of interest, and obtain a second local superpixel region set; calculate the eigenvalue of the second local superpixel region set in the second target cropping band corresponding to the second region of interest, and obtain a second eigenvalue; calculate the second eigenvalue and the second target cropping band corresponding to the second region of interest, and obtain the edge detail nuisance coefficient of the second region of interest; and calculate each region of interest in turn to obtain the edge detail nuisance coefficient corresponding to each region of interest.

[0128] In one embodiment, in determining the cropping scale corresponding to each region of interest according to the edge detail disturbance coefficient corresponding to each region of interest and the edge diffusion coefficient corresponding to each region of interest, the determination unit 206 is further used to: obtain the edge detail loss rate corresponding to each region of interest according to the edge detail disturbance coefficient corresponding to each region of interest and the edge diffusion coefficient corresponding to each region of interest; obtain the detail feature loss factor corresponding to each region of interest according to the edge detail loss rate corresponding to each region of interest and the corresponding interest degree coefficient carried by each region of interest; and determine the cropping scale corresponding to each region of interest according to the detail feature loss factor corresponding to each region of interest.

[0129] In one embodiment, in the acquisition of the edge diffusion coefficient corresponding to each region of interest, the acquisition unit 205 is further used to: acquire a third region of interest, a third remote sensing image of a first band where the third region of interest is located, and a fourth remote sensing image of a second band where the third region of interest is located, wherein the third region of interest is any region of interest among each region of interest, and the first band and the second band are adjacent bands; in the third remote sensing image, all first edge curve pixel point coordinates and first centroid coordinates corresponding to the third region of interest are calculated to obtain a first distance; in the fourth remote sensing image, all second edge curve pixel point coordinates and second centroid coordinates corresponding to the third region of interest are calculated to obtain a second distance; the first distance and the second distance are calculated to obtain the edge diffusion coefficient corresponding to the third region of interest; and each region of interest is calculated in turn to obtain the edge diffusion coefficient corresponding to each region of interest.

[0130] It should be noted that the above-mentioned remote sensing image cropping device based on multi-band remote sensing images can execute the remote sensing image cropping method based on multi-band remote sensing images provided in the embodiment of the present application, and has the corresponding functional modules and beneficial effects of the execution method. For technical details not fully described in the embodiment of the remote sensing image cropping device based on multi-band remote sensing images, please refer to the remote sensing image cropping method based on multi-band remote sensing images provided in the embodiment of the present application.

[0131] See also Figure 3 , Figure 3 Schematic diagram of a remote sensing image cropping system based on multi-band remote sensing images provided in an embodiment of the present application. Figure 3 As shown, the system 300 includes a processor 301 and a memory 302. The processor 301 is in communication connection with the memory 302.

[0132] The processor 301 is configured to support the system to execute the corresponding functions in the remote sensing image cropping method based on multi-band remote sensing images in the above method embodiment. The processor 301 can be a central processing unit (CPU), a network processor (NP), a hardware chip or any combination thereof. The above hardware chip can be an application specific integrated circuit (ASIC), a programmable logic device (PLD) or a combination thereof. The above PLD can be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL) or any combination thereof.

[0133] Specifically, the processor 301 may include a sending card, a receiving card and a driving chip.

[0134] The memory 302 is used to store program codes, etc. The memory 302 may include a volatile memory (VM), such as a random access memory (RAM); the memory 302 may also include a non-volatile memory (NVM), such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD); the memory 302 may also include a combination of the above-mentioned types of memories.

[0135] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer executes the remote sensing image cropping method based on multi-band remote sensing images as described in the above-mentioned embodiment.

[0136] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

Claims

1. A remote sensing image cropping method based on multi-band remote sensing images, characterized in that: Applied to a remote sensing image cropping system based on multi-band remote sensing images, the method comprises: Acquire initial remote sensing images of different bands, perform region segmentation processing on the initial remote sensing images, and obtain multiple local super-pixel regions with similar pixel features; Performing region recognition on the multiple local superpixel regions to obtain at least one region of interest; Performing edge detection on the multiple local super-pixel regions to obtain an edge curve corresponding to each local super-pixel region; Performing regional analysis on the at least one region of interest according to the edge curve corresponding to each local superpixel region to obtain an edge detail nuisance coefficient corresponding to each region of interest; Obtaining an edge diffusion coefficient corresponding to each region of interest, where the edge diffusion coefficient is an expansion coefficient in a same local superpixel region corresponding to each region of interest in different bands; Determining a cropping scale corresponding to each region of interest according to an edge detail nuisance coefficient corresponding to each region of interest and an edge diffusion coefficient corresponding to each region of interest; Cropping each region of interest according to the cropping scale corresponding to each region of interest to obtain each cropped region of interest; Each cropped region of interest is spliced ​​to obtain a target remote sensing image.

2. The remote sensing image cropping method based on multi-band remote sensing images according to claim 1 is characterized in that: The performing region recognition on the multiple local superpixel regions to obtain at least one region of interest includes: Performing corner point detection processing on the multiple local superpixel regions to obtain all feature corner points in the multiple local superpixel regions; Calculating based on all feature corner points in the multiple local superpixel regions, obtaining an interest degree coefficient corresponding to each local superpixel region; The interest level coefficient corresponding to each local superpixel region is compared with a preset interest level threshold, and local superpixel regions with a level greater than the preset interest level threshold are screened out as interest regions, thereby obtaining at least one interest region.

3. The remote sensing image cropping method based on multi-band remote sensing images according to claim 2 is characterized in that: The calculating according to all the feature corner points in the plurality of local superpixel regions to obtain the interest degree coefficient corresponding to each local superpixel region includes: Acquire a target local superpixel region and a first corner point set corresponding to the target local superpixel region, wherein the target local superpixel region is any region among the multiple local superpixel regions, and the first corner point set includes all corner points in the target local superpixel region; Acquire a first remote sensing image of the target local superpixel region at a first moment and a second remote sensing image at a second moment, wherein the second moment is a moment adjacent to the left of the first moment; Acquire a first target local superpixel corner point set corresponding to the first remote sensing image and a second target local superpixel corner point set corresponding to the second remote sensing image; Obtaining a corner point difference between the target local superpixel region in the first remote sensing image and the second remote sensing image according to the first target local superpixel corner point set and the second target local superpixel corner point set; Performing distance calculation based on the first corner point set corresponding to the target local superpixel region, obtaining the mean Euclidean distance between all corner points in the target local superpixel region; Calculating the mean of the Euclidean distances between all corner points in the target local superpixel region and the corner point difference to obtain the interest degree coefficient corresponding to the target local superpixel region; Calculate each local superpixel region in turn to obtain the interest degree coefficient corresponding to each local superpixel region.

4. The remote sensing image cropping method based on multi-band remote sensing images according to claim 1, characterized in that: The performing of regional analysis on the at least one region of interest according to the edge curve corresponding to each local superpixel region to obtain the edge detail nuisance coefficient corresponding to each region of interest includes: Performing feature analysis on the at least one region of interest according to the edge curve corresponding to each local superpixel region to obtain a feature value corresponding to each region of interest; Determine, according to the characteristic value corresponding to each region of interest, a target cropping band corresponding to each region of interest in different bands; According to the target cropping band corresponding to each region of interest, an edge detail nuisance coefficient corresponding to each region of interest is obtained.

5. The remote sensing image cropping method based on multi-band remote sensing images according to claim 4 is characterized in that: The performing feature analysis on the at least one region of interest according to the edge curve corresponding to each local superpixel region to obtain a feature value corresponding to each region of interest includes: Acquire a target region of interest and a target edge curve where the target region of interest is located, wherein the target region of interest is any region of the at least one region of interest; Acquire the total number of pixels in the target interest region and the number of edge pixels on the target edge curve where the target interest region is located; Obtaining an edge density coefficient of the target interest region according to the total number of pixels in the target interest region and the number of edge pixels on the target edge curve where the target interest region is located; Acquire all local superpixel regions adjacent to the target region of interest to obtain a first local superpixel region set; Calculate the total number of pixels in the target area of ​​interest to obtain a mean pixel value corresponding to the target area of ​​interest; Obtaining an interest degree coefficient corresponding to each first local superpixel region in the first local superpixel region set; Calculating the edge density coefficient of the target interest region, the mean pixel value corresponding to the target interest region, and the interest degree coefficient corresponding to each first local superpixel region to obtain a feature value corresponding to the target interest region; Calculate each region of interest in turn to obtain a feature value corresponding to each region of interest.

6. The remote sensing image cropping method based on multi-band remote sensing images according to claim 4, characterized in that: The determining, according to the characteristic value corresponding to each region of interest, a target cropping band corresponding to each region of interest in different bands includes: Acquire a plurality of first characteristic values ​​of a first region of interest in different bands, where the first region of interest is any region of interest in each of the regions of interest; Arranging the plurality of first eigenvalues ​​according to a preset arrangement rule to obtain a target sequence; Selecting the first eigenvalue with the largest value in the target sequence as the target eigenvalue; Determine that the initial remote sensing image corresponding to the target characteristic value is the first target cropping band corresponding to the first area of ​​interest; Band processing is performed on each of the regions of interest in turn to obtain a target cropped band corresponding to each region of interest.

7. The remote sensing image cropping method based on multi-band remote sensing images according to claim 4, characterized in that: The step of obtaining the edge detail nuisance coefficient corresponding to each region of interest according to the target cropping band corresponding to each region of interest includes: Acquire a second region of interest and a second target clipping band corresponding to the second region of interest, where the second region of interest is any region of interest in each of the regions of interest; Acquire all local superpixel regions adjacent to the second region of interest to obtain a second local superpixel region set; Calculating the eigenvalue of the second local superpixel region set in the second target cropping band corresponding to the second region of interest to obtain a second eigenvalue; Calculating the second eigenvalue and the second target cropped band corresponding to the second region of interest to obtain an edge detail nuisance coefficient of the second region of interest; Calculate each of the regions of interest in turn to obtain an edge detail nuisance coefficient corresponding to each region of interest.

8. The remote sensing image cropping method based on multi-band remote sensing images according to claim 1, characterized in that: The determining the cropping scale corresponding to each region of interest according to the edge detail nuisance coefficient corresponding to each region of interest and the edge diffusion coefficient corresponding to each region of interest includes: Obtaining an edge detail loss rate corresponding to each region of interest according to an edge detail annoyance coefficient corresponding to each region of interest and an edge diffusion coefficient corresponding to each region of interest; According to the edge detail loss rate corresponding to each region of interest and the interest degree coefficient corresponding to each region of interest, a detail feature loss factor corresponding to each region of interest is obtained; The cropping scale corresponding to each region of interest is determined according to the detail feature loss factor corresponding to each region of interest.

9. The remote sensing image cropping method based on multi-band remote sensing images according to claim 1, characterized in that: The obtaining of the edge diffusion coefficient corresponding to each region of interest includes: Acquire a third region of interest, a third remote sensing image of a first band where the third region of interest is located, and a fourth remote sensing image of a second band where the third region of interest is located, wherein the third region of interest is any region of interest among each of the regions of interest, and the first band and the second band are adjacent bands; In the third remote sensing image, the coordinates of all first edge curve pixel points and the first centroid coordinates corresponding to the third region of interest are calculated to obtain a first distance; In the fourth remote sensing image, the coordinates of all second edge curve pixel points and the second centroid coordinates corresponding to the third region of interest are calculated to obtain a second distance; Calculating the first distance and the second distance to obtain an edge diffusion coefficient corresponding to the third region of interest; Calculate each region of interest in turn to obtain an edge diffusion coefficient corresponding to each region of interest.

10. A remote sensing image cropping system based on multi-band remote sensing images, characterized in that: The system comprises a memory and a processor, wherein the memory is connected to the processor, and the processor is used to execute one or more computer programs stored in the memory. When the processor executes the one or more computer programs, the remote sensing image cropping system based on multi-band remote sensing images implements the remote sensing image cropping method based on multi-band remote sensing images as described in any one of claims 1 to 9.

Images