Remote Sensing Image Target Detection Method, Device and Medium Based on Foreground Super-Resolution

Abstract

The present invention discloses a remote sensing image target detection method, device and medium based on foreground super-resolution. The present invention performs target detection processing on a first image block set and a third image block set of two channels with different resolutions, which can be applicable to the recognition of targets of different sizes, and both large and small targets can be recognized, thereby improving practicability. The second image block set for super-resolution reconstruction processing is previously subjected to foreground segmentation processing, which can eliminate the invalid background of the remote sensing image and extract the area containing the target, so that the subsequent super-resolution reconstruction processing can be focused on the foreground that needs to be paid attention to, avoiding the calculation cost from being wasted on the invalid background, thereby improving the efficiency of the super-resolution reconstruction processing, and then improving the target detection performance, with high practicability and high efficiency. As a remote sensing image target detection method, device and medium based on foreground super-resolution, the present invention can be widely used in the field of image processing technology.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method, device and medium for remote sensing image target detection based on foreground super-resolution. Background Art

[0002] In remote sensing image processing, the input data is aerospace or aerial remote sensing images, and the image size is usually very large, reaching the level of tens of thousands of pixels by tens of thousands of pixels. Therefore, it is impossible to directly apply the target detection model to remote sensing images, and the existing target detection models cannot be directly applied to the target detection of remote sensing images. At the same time, remote sensing image targets often have the following characteristics: uneven distribution of targets, large differences in target sizes, and low image resolution. Based on the characteristics of remote sensing image targets, it shows that general target detection algorithms will inevitably spend a large amount of computing costs in the background area without targets, resulting in waste of resources and low detection efficiency. Moreover, it will be limited by low resolution. Even if the dense detection method is adopted, the detection of small targets is still ineffective, that is, the existing detection methods cannot effectively detect small targets. Therefore, it is necessary to research and design corresponding detection methods to solve this problem. Summary of the Invention

[0003] In view of this, in order to solve the above technical problems, the purpose of the present invention is to provide a method, device and medium for remote sensing image target detection based on foreground super-resolution, so as to improve the detection efficiency and practicality.

[0004] The technical solution adopted by the present invention is as follows:

[0005] A method for remote sensing image target detection based on foreground super-resolution, comprising:

[0006] Obtain a remote sensing image;

[0007] Perform image block processing on the remote sensing image to obtain a first image block set;

[0008] Perform foreground segmentation processing on the first image block set to obtain a second image block set; the second image block set includes a plurality of foreground small cut blocks containing targets;

[0009] Perform super-resolution reconstruction processing on the second image block set to obtain a third image block set;

[0010] Perform first target detection processing on the first image block set and second target detection processing on the third image block set; the first target detection processing and the second target detection processing are used to detect targets of different sizes;

[0011] According to the first target detection processing result and the second target detection processing result, perform non-maximum suppression to determine the target detection result.

[0012] Further, the image segmentation process for the remote sensing image to obtain the first image block set includes:

[0013] Determine the initial area in the remote sensing image according to the preset segmentation order and the preset segmentation scale;

[0014] Perform a segmentation process on the initial area, and move from the initial area according to the preset segmentation order and the preset segmentation scale and perform a segmentation process during the movement to determine a plurality of first image blocks; wherein the overlapping rates of adjacent first image blocks in the horizontal and vertical directions are both greater than or equal to the first preset overlapping rate.

[0015] Further, the first image block set includes a plurality of the first image blocks, and the foreground segmentation process for the first image block set to obtain the second image block set includes:

[0016] Determine the target distribution density map of the first image block through the foreground segmentation model;

[0017] According to the target distribution density map, use the preset scale to perform a cropping process on the first image block to obtain the second image block set; wherein the preset scale is smaller than the preset segmentation scale, and the overlapping rates of adjacent foreground small cut blocks corresponding to the same first image block in the horizontal and vertical directions are both greater than or equal to the second preset overlapping rate.

[0018] Further, the super-resolution reconstruction process for the second image block set to obtain the third image block set includes:

[0019] Magnify the pixels of the foreground small cut blocks by a preset multiple through the super-resolution reconstruction model, and perform a detail reconstruction process on the foreground small cut blocks magnified by the preset multiple to obtain the third image block set.

[0020] Further, the first target detection process for the first image block set includes:

[0021] Perform a first target detection process on the first image block set through the target detection model to identify the target of the first size.

[0022] Further, the second target detection process for the third image block set includes:

[0023] Perform a second target detection process on the third image block set through the target detection model to identify the target of the second size; the second size is smaller than the first size.

[0024] Further, the first object detection processing result includes first prediction information, and the second object detection processing result includes second prediction information. Determining the object detection result through non-maximum suppression based on the first object detection processing result and the second object detection processing result includes:

[0025] Adjust the second prediction information so that the second prediction information is adjusted to the position representation in the remote sensing image;

[0026] Pool the adjusted second prediction information with the first prediction information;

[0027] Perform non-maximum suppression on the pooling result to determine the object detection result.

[0028] The present invention also provides a remote sensing image object detection device based on foreground super-resolution, including:

[0029] An acquisition module, configured to acquire a remote sensing image;

[0030] A chunking module, configured to perform image chunking processing on the remote sensing image to obtain a first set of image chunks;

[0031] A foreground segmentation module, configured to perform foreground segmentation processing on the first set of image chunks to obtain a second set of image chunks; the second set of image chunks includes a plurality of foreground small chunks containing objects;

[0032] A super-resolution reconstruction module, configured to perform super-resolution reconstruction processing on the second set of image chunks to obtain a third set of image chunks;

[0033] An object detection module, configured to perform first object detection processing on the first set of image chunks and second object detection processing on the third set of image chunks; the first object detection processing and the second object detection processing are used to detect objects of different sizes;

[0034] A determination module, configured to perform non-maximum suppression based on the first object detection processing result and the second object detection processing result to determine the object detection result.

[0035] The present invention also provides a remote sensing image object detection device based on foreground super-resolution, including a processor and a memory;

[0036] The memory stores a program;

[0037] The processor executes the program to implement the method.

[0038] The present invention also provides a computer-readable storage medium, where the storage medium stores a program, and when the program is executed by a processor, the method is implemented.

[0039] The beneficial effects of the present invention are as follows: By obtaining a remote sensing image, performing image block processing on the remote sensing image to obtain a first set of image blocks, performing foreground segmentation processing on the first set of image blocks to obtain a second set of image blocks, performing super-resolution reconstruction processing on the second set of image blocks to obtain a third set of image blocks, performing first target detection processing on the first set of image blocks, and performing second target detection processing on the third set of image blocks; the first target detection processing and the second target detection processing are used to detect targets of different sizes; according to the first target detection processing result and the second target detection processing result, non-maximum suppression is performed to determine the target detection result; by performing target detection processing on two sets of first image blocks and third image blocks with different resolutions, it can be applied to the recognition of targets of different sizes, both large and small targets can be recognized, improving the practicability; and the second set of image blocks subjected to super-resolution reconstruction processing has been previously subjected to foreground segmentation processing, which can eliminate the invalid background of the remote sensing image and extract the area containing the target, so that subsequent super-resolution reconstruction processing can focus on the foreground that needs attention, avoiding waste of computational cost in the invalid background, thereby improving the efficiency of super-resolution reconstruction processing, and further improving the target detection performance, with high practicability and high efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic flowchart of the steps of the method for detecting targets in remote sensing images based on foreground super-resolution of the present invention;

[0041] Figure 2 It is a schematic diagram of the steps for determining the first set of image blocks in a specific embodiment of the present invention;

[0042] FIG. 3(a) is the first schematic diagram of block processing in a specific embodiment of the present invention, and FIG. 3(b) is the second schematic diagram of block processing in a specific embodiment of the present invention;

[0043] FIG. 4(a) is the effect diagram of the existing detection method, and FIG. 4(b) is the effect diagram of the target detection result in a specific embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] In order to enable those skilled in the art of the present technology to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0045] In the description, claims and drawings of this application, terms such as "first", "second", "third" and "fourth" are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally further include steps or units not listed, or may optionally further include other steps or units inherent to these processes, methods, products or devices.

[0046] Reference to "embodiment" herein means that a specific feature, structure or characteristic described in connection with the embodiment may be included in at least one embodiment of this application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein may be combined with other embodiments.

[0047] As Figure 1 shown, an embodiment of the present invention provides a remote sensing image target detection method based on foreground super-resolution, including steps S100 - S600:

[0048] S100. Obtain a remote sensing image.

[0049] In an embodiment of the present invention, the remote sensing image can be obtained through a remote sensing sensing device or a remote sensing equipment, or can also be downloaded from a server, a storage device or the Internet, without specific limitation.

[0050] S200. Perform image block processing on the remote sensing image to obtain a first image block set.

[0051] In an embodiment of the present invention, the first image block set includes a plurality of first image blocks.

[0052] As Figure 2 shown, specifically, step S200 includes steps S210 - S220:

[0053] S210. Determine an initial area in the remote sensing image according to a preset block sequence and a preset block scale.

[0054] In an embodiment of the present invention, the preset block sequence can be from left to right or from right to left horizontally, and from top to bottom or from bottom to top vertically. The preset block scale can be set according to actual needs, without specific limitation. For example, taking the preset block sequence from left to right and from top to bottom, and the preset block scale of 1024*1024 as an example, the upper left corner of the remote sensing image can be selected according to the preset block sequence from left to right and from top to bottom, and an initial area with a scale of 1024*1024 can be determined at the upper left corner.

[0055] S220. Perform a chunking process on the initial region, and move from the initial region according to a preset chunking order and a preset chunking scale, and perform a chunking process during the movement to determine a plurality of first image chunks.

[0056] In the embodiments of the present invention, taking the above example, the initial region with a scale of 1024*1024 is chunked into the first first image chunk, and taking this initial region as the starting point, first from left to right and then from top to bottom, or first from top to bottom and then from left to right, other regions with a scale of 1024*1024 are determined one by one until all regions of the entire remote sensing image are obtained, obtaining a plurality of first image chunks, and the scales of the first image chunks are the same, all being the preset chunking scale, that is, 1024*1024. Specifically, if the range of the first image chunk exceeds the boundary of the entire remote sensing image, such as the horizontal boundary or the vertical boundary, the region of 1024*1024 is automatically moved a certain number of pixel distances in the direction away from the boundary until the edge of the region of 1024*1024 is within the boundary of the remote sensing image. It should be noted that in order to ensure that some first image chunks in the image chunking process can cover more information of the target (the target to be detected) without over-segmenting the target, in the embodiments of the present invention, the overlapping rates of adjacent first image chunks in the horizontal and vertical directions are both greater than or equal to the first preset overlapping rate. Optionally, the first preset overlapping rate can be set as needed. In the embodiments of the present invention, taking the first preset overlapping rate as 50% as an example, as shown in FIG. 3(a), the initial region, that is, the first first image chunk 301, when moving to the right from the first image chunk 301, at this time, the overlapping rates of the adjacent first image chunk 302 with the first image chunk 301 in the horizontal and vertical directions are both greater than or equal to 50%; as shown in FIG. 3(b), when moving down from the first image chunk 301, at this time, the overlapping rates of the adjacent first image chunk 303 with the first image chunk 301 in the horizontal and vertical directions are both greater than or equal to 50%.

[0057] S300. Perform a foreground segmentation process on the set of first image chunks to obtain a set of second image chunks.

[0058] In the embodiments of the present invention, the set of second image chunks includes a plurality of foreground small chunks containing the target.

[0059] Specifically, step S300 includes steps S310 - S320:

[0060] S310. Determine the target distribution density map of the first image chunk through the foreground segmentation model.

[0061] In the embodiments of the present invention, the foreground segmentation model can be pre-trained and stored. Optionally, the foreground segmentation model can be a model based on the MCNN density map estimation algorithm. Through the learning ability of the neural network, for each first image block input at a time, the corresponding target distribution density map of each first image block can be obtained.

[0062] S320. According to the target distribution density map, use a preset scale to perform cropping processing on the first image block to obtain a second set of image blocks.

[0063] In the embodiments of the present invention, the preset scale is smaller than the preset block scale and is used to select small-sized targets. The preset scale can also be adjusted according to actual needs and is not specifically limited. In the embodiments of the present invention, the preset scale of 256*256 is taken as an example for illustration. Similarly, in order to ensure that some of the second image blocks obtained by the cropping processing can cover more information of the target (the target to be detected), the overlap rate of adjacent foreground small cut blocks corresponding to the same first image block in the horizontal and vertical directions is greater than or equal to the second preset overlap rate. Similarly, when performing the cropping processing, the preset scale can be combined according to the preset order. For example, when performing the cropping processing on the initial area, that is, the first first image block, multiple foreground small cut blocks with a preset scale of 256*256 are determined according to the preset order. When performing the cropping processing on this first image block, the overlap rate of the corresponding adjacent foreground small cut blocks in the horizontal and vertical directions is greater than or equal to the second preset overlap rate. By performing the cropping processing on each first image block, the foreground small cut blocks corresponding to all first image blocks can be obtained. It should be noted that the second preset overlap rate can be adjusted according to needs, and can be the same as the first preset overlap rate, such as 50%, or can be different from the first preset overlap rate. In addition, in the embodiments of the present invention, multiple foreground small cut blocks containing the target and having the same size as the preset scale are obtained through the foreground segmentation model. The foreground small cut blocks do not contain target detection-related information such as the category and position box of the target. Its function is to extract the area containing the target by removing the invalid background of the remote sensing image, so that the subsequent super-resolution reconstruction processing can focus on the foreground that needs to be concerned, avoid wasting the computing cost on the invalid background, thereby improving the efficiency of the super-resolution reconstruction processing, and further improving the target detection performance, with high practicality and high efficiency.

[0064] S400. Perform super-resolution reconstruction processing on the second set of image blocks to obtain a third set of image blocks.

[0065] In the embodiments of the present invention, the third set of image blocks includes multiple third image blocks. Among them, the super-resolution reconstruction processing refers to the process of magnifying a low-resolution image according to a certain super-resolution ratio and obtaining a high-resolution image through detail reconstruction.

[0066] Specifically, step S400 includes step S410:

[0067] S410. Use a super-resolution reconstruction model to magnify the pixels of the foreground small cut by a preset multiple, and perform detail reconstruction processing on the foreground small cut magnified by the preset multiple to obtain a third set of image blocks.

[0068] In the embodiment of the present invention, the super-resolution reconstruction model can be a pre-trained model, including but not limited to being trained by the ESRGAN algorithm. Among them, the preset multiple can be adjusted as needed. In the embodiment of the present invention, taking 4 times as an example, that is, magnify the pixels of each foreground small cut by 4 times, from 256*256 to 1024*1024, and perform detail reconstruction processing on the foreground small cut magnified by 4 times to obtain the third image block corresponding to the foreground small cut, constituting the third set of image blocks. It should be noted that since the pixels of each foreground small cut are magnified by 4 times, the magnified foreground small cut will be blurred. Therefore, through detail reconstruction processing, the blurred foreground small cut can be made clear and the resolution can be improved.

[0069] S500. Perform a first target detection process on the first set of image blocks and a second target detection process on the third set of image blocks.

[0070] In the embodiment of the present invention, the first target detection process and the second target detection process are used to detect targets of different sizes.

[0071] Specifically, the first target detection process on the first set of image blocks in step S500 includes step S510:

[0072] S510. Use a target detection model to perform a first target detection process on the first set of image blocks to identify targets of the first size.

[0073] Similarly, the target detection model can be obtained through pre-training. In the embodiment of the present invention, after the first target detection process, a first target detection result can be obtained. The first target detection result contains first prediction information, and the first prediction information includes information related to targets of the first size, including but not limited to position coordinates X1, Y1, length, width, rotation angle, category, accuracy rate attribute, etc. It should be noted that the targets of the first size can be medium and large-sized targets. For example, a preset standard can be set. When the target meets the preset standard, it is a medium and large-sized target, otherwise it is a small-sized target. The medium and large-sized targets and small-sized targets are relative in a remote sensing image.

[0074] Specifically, the second target detection process on the third set of image blocks in step S500 includes step S520, where the execution order of step S510 and step S520 is not limited:

[0075] S520. Perform a second object detection process on the third set of image patches through an object detection model to identify objects of a second size; the second size is smaller than the first size.

[0076] Similarly, the object detection model can be obtained through pre-training. In the embodiments of the present invention, the same object detection model is used for the first object detection process and the second object detection process. In other embodiments, different object detection models can be used, which is not specifically limited. Among them, the object detection model can identify objects of the second size in each third image patch. It should be noted that after the second object detection process, a second object detection result can be obtained. The second object detection result contains second prediction information, and the second prediction information includes information related to objects of the second size, including but not limited to relative position coordinates X2, Y2, length, width, rotation angle, category, accuracy attribute, etc. It should be noted that the second size is smaller than the first size, and the objects of the second size can be small objects. Similarly, a preset standard can be set. When the object does not meet the preset standard, it is a small object; when it meets the standard, it is a medium or large object. The medium or large objects and small objects are relative in a remote sensing image.

[0077] S600. According to the first object detection result and the second object detection result, perform non-maximum suppression to determine the object detection result.

[0078] Specifically, step S600 includes steps S610 - S630:

[0079] S610. Adjust the second prediction information so that the second prediction information is adjusted to a position representation in the remote sensing image.

[0080] In the embodiments of the present invention, since the position of the object in the second prediction information is the relative position coordinates X2, Y2, that is, the position of the object in the foreground small patch or the third image patch, it is necessary to combine the position of the foreground small patch in the original remote sensing image to adjust the relative position to an absolute position, so that the second prediction information is adjusted to a position representation in the remote sensing image, that is, the relative position coordinates X2, Y2 are adjusted to a position representation in the remote sensing image.

[0081] S620. Pool the adjusted second prediction information and the first prediction information.

[0082] In the embodiments of the present invention, the adjusted second prediction information and the first prediction information are pooled, for example, pooled into the remote sensing image for the next step of processing.

[0083] S630. Perform non-maximum suppression on the pooling result to determine the object detection result.

[0084] In the embodiments of the present invention, the target detection processing results of two paths are pooled together and regarded as the output detection results of a target detection model. Then, non-maximum suppression is performed on the pooled results to remove redundant detection frames, and the final target detection results are obtained.

[0085] The remote sensing image target detection method based on foreground super-resolution in the embodiments of the present invention uses target detection on two images with different resolutions, namely the first image block set and the third image block set, to achieve the detection and recognition of large, medium, and small targets. Among them, due to the low resolution of the first image block set, the information of large and medium-sized targets is retained, and the detection performance for small targets is relatively poor. On the other hand, the third image block set can improve the detection performance for small targets after foreground segmentation processing and super-resolution reconstruction processing. However, due to the foreground cutting processing involved in the process, the original shape of large targets is segmented, so the detection of large targets is not good. By using two inputs with different resolutions, each focusing on targets of different scales, the two are combined to achieve the detection and recognition of large, medium, and small targets.

[0086] As shown in FIG. 4(a), the detection result of the existing detection method can only detect large and medium-sized targets 401 and cannot effectively detect small targets. As shown in FIG. 4(b), the target detection result of the remote sensing image target detection method based on foreground super-resolution in the embodiments of the present invention can not only detect large and medium-sized targets 401 but also effectively detect small targets 402, with good recognition effect and strong practicability.

[0087] The embodiments of the present invention also provide a remote sensing image target detection device based on foreground super-resolution, including:

[0088] An acquisition module for acquiring remote sensing images;

[0089] A block division module for performing image block division processing on the remote sensing image to obtain a first image block set;

[0090] A foreground segmentation module for performing foreground segmentation processing on the first image block set to obtain a second image block set; the second image block set includes multiple foreground small cut blocks containing targets;

[0091] A super-resolution reconstruction module for performing super-resolution reconstruction processing on the second image block set to obtain a third image block set;

[0092] A target detection module for performing first target detection processing on the first image block set and second target detection processing on the third image block set; the first target detection processing and the second target detection processing are used to detect targets of different sizes;

[0093] A determination module for performing non-maximum suppression according to the first target detection processing result and the second target detection processing result to determine the target detection result.

[0094] The content in the above method embodiments is applicable to the device embodiments herein. The functions specifically implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those of the above method embodiments.

[0095] An embodiment of the present invention further provides a remote sensing image target detection device based on foreground super-resolution. The device includes a processor and a memory;

[0096] The memory is used to store programs;

[0097] The processor is used to execute the program to implement the remote sensing image target detection method based on foreground super-resolution in the embodiment of the present invention. The device in the embodiment of the present invention can implement the function of remote sensing image target detection based on foreground super-resolution. The device can be any intelligent terminal including a mobile phone, a tablet computer, a computer, a personal digital assistant (PDA for short), a point of sales (POS for short), an in-vehicle computer, etc.

[0098] The content in the above method embodiments is applicable to the device embodiments herein. The functions specifically implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those of the above method embodiments.

[0099] An embodiment of the present invention further provides a computer-readable storage medium. The computer-readable storage medium stores a program, and the program is executed by the processor to complete the remote sensing image target detection method based on foreground super-resolution in the foregoing embodiment of the invention.

[0100] An embodiment of the present invention further provides a computer program product including instructions. When it runs on a computer, it causes the computer to execute the remote sensing image target detection method based on foreground super-resolution in the foregoing embodiment of the invention.

[0101] Terms such as "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments of the present application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0102] It should be understood that in this application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the relationship between associated objects and indicates that there can be three relationships. For example, "A and / or B" can mean: only A exists, only B exists, and both A and B exist simultaneously. Here, A and B can be singular or plural. The character " / " generally indicates an "or" relationship between the associated objects before and after. "At least one (of the following)" or its similar expressions refer to any combination of these items, including any combination of single items or plural items. For example, at least one (of) a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0103] In several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling, direct coupling, or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of devices or units can be in electrical, mechanical, or other forms. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in each embodiment of this application, the functional units can be integrated in one processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0104] When an integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of various embodiments of this application. The foregoing storage medium includes: various media that can store programs, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0105] The above, the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of various embodiments of this application.

Claims

1. A method for remote sensing image target detection based on foreground super-resolution, characterized in that, Including: Obtaining remote sensing images; Performing image block processing on the remote sensing images to obtain a first set of image blocks; Performing foreground segmentation processing on the first set of image blocks to obtain a second set of image blocks; the second set of image blocks includes multiple foreground small cut blocks containing targets; Performing super-resolution reconstruction processing on the second set of image blocks to obtain a third set of image blocks; Performing first target detection processing on the first set of image blocks and performing second target detection processing on the third set of image blocks; The first target detection processing and the second target detection processing are used to detect targets of different sizes; According to the first target detection processing result and the second target detection processing result, performing non-maximum suppression to determine the target detection result; The first set of image blocks includes multiple of the first image blocks, and the performing foreground segmentation processing on the first set of image blocks to obtain a second set of image blocks includes: Determining the target distribution density map of the first image block through a foreground segmentation model; According to the target distribution density map, using a preset scale to perform cropping processing on the first image block to obtain the second set of image blocks; wherein the preset scale is smaller than the preset block scale, and the overlapping rate of adjacent foreground small cut blocks corresponding to the same first image block in the horizontal and vertical directions is greater than or equal to a second preset overlapping rate; The performing first target detection processing on the first set of image blocks includes: Performing first target detection processing on the first set of image blocks through a target detection model to identify targets of a first size; The performing second target detection processing on the third set of image blocks includes: Performing second target detection processing on the third set of image blocks through a target detection model to identify targets of a second size; the second size is smaller than the first size; The first target detection processing result includes first prediction information, the second target detection processing result includes second prediction information, and the according to the first target detection processing result and the second target detection processing result, performing non-maximum suppression to determine the target detection result includes: Adjusting the second prediction information so that the second prediction information is adjusted to a position representation in the remote sensing image; Pooling the adjusted second prediction information with the first prediction information; Performing non-maximum suppression on the pooling result to determine the target detection result.

2. The method for remote sensing image target detection based on foreground super-resolution according to claim 1, characterized in that: The performing image block processing on the remote sensing images to obtain a first set of image blocks includes: Determining an initial area in the remote sensing image according to a preset block order and a preset block scale; Performing block processing on the initial area, and moving from the initial area according to the preset block order and the preset block scale and performing block processing during the movement to determine multiple first image blocks; wherein the overlapping rate of adjacent first image blocks in the horizontal and vertical directions is greater than or equal to a first preset overlapping rate.

3. The method for remote sensing image target detection based on foreground super-resolution according to claim 1, characterized in that: The performing super-resolution reconstruction processing on the second set of image blocks to obtain a third set of image blocks includes: The pixels of the foreground small cut are magnified by a preset multiple through a super-resolution reconstruction model, and the foreground small cut magnified by the preset multiple is subjected to detail reconstruction processing to obtain a third set of image blocks.

4. A device for remote sensing image target detection based on foreground super-resolution, characterized in that, It includes: An acquisition module for acquiring remote sensing images; A block division module for performing image block division processing on the remote sensing image to obtain a first set of image blocks; A foreground segmentation module for performing foreground segmentation processing on the first set of image blocks to obtain a second set of image blocks; the second set of image blocks includes multiple foreground small cuts containing targets; A super-resolution reconstruction module for performing super-resolution reconstruction processing on the second set of image blocks to obtain a third set of image blocks; A target detection module for performing first target detection processing on the first set of image blocks and second target detection processing on the third set of image blocks; The first target detection processing and the second target detection processing are used to detect targets of different sizes; A determination module for performing non-maximum suppression according to the first target detection processing result and the second target detection processing result to determine the target detection result; The first set of image blocks includes multiple first image blocks, and the performing foreground segmentation processing on the first set of image blocks to obtain a second set of image blocks includes: Determining the target distribution density map of the first image block through a foreground segmentation model; According to the target distribution density map, performing cropping processing on the first image block by using a preset scale to obtain the second set of image blocks; wherein the preset scale is smaller than the preset block division scale, and the overlapping rate of adjacent foreground small cuts corresponding to the same first image block in the horizontal and vertical directions is greater than or equal to a second preset overlapping rate; The performing first target detection processing on the first set of image blocks includes: Performing first target detection processing on the first set of image blocks through a target detection model to identify targets of a first size; The performing second target detection processing on the third set of image blocks includes: Performing second target detection processing on the third set of image blocks through a target detection model to identify targets of a second size; the second size is smaller than the first size; The first target detection processing result includes first prediction information, the second target detection processing result includes second prediction information, and the performing non-maximum suppression according to the first target detection processing result and the second target detection processing result to determine the target detection result includes: Adjusting the second prediction information so that the second prediction information is adjusted to the position representation in the remote sensing image; Pooling the adjusted second prediction information and the first prediction information; Performing non-maximum suppression on the pooling result to determine the target detection result.

5. A remote sensing image target detection device based on foreground super-resolution, characterized in that, It includes a processor and a memory; The memory stores a program; The processor executes the program to implement the method according to any one of claims 1-3.

6. A computer-readable storage medium, characterized in that, The storage medium stores a program, and when the program is executed by the processor, the method according to any one of claims 1-3 is implemented.

Images