High-value target rapid detection and positioning method and system based on ultra-wide rotary scanning remote sensing image

Abstract

The invention discloses a high-value target rapid detection and positioning method and system based on an ultra-wide rotary scanning remote sensing image, belongs to the technical field of computer algorithms, and is used for realizing high-value target rapid identification and positioning under the remote sensing image. The method comprises the following steps: acquiring an ultra-wide swing-scan remote sensing image, and timestamps, remote sensing satellite orbit data and swing-scan camera parameters corresponding to each column of pixels of the image; according to the data, calculating an overlapping proportion required by a corresponding pixel position in real time, and segmenting the ultra-wide swing-scanning remote sensing image into a plurality of slices with different sizes; taking a plurality of slices with different sizes as the input of a detector, outputting a target detection frame, and mapping the detection frame back to the ultra-wide remote sensing image according to the positions of the slices in the original image; and taking the pixel coordinate of the center of the target detection frame as the image position of the target, and mapping the pixel coordinate of the target to a WGS84 coordinate to complete the detection and positioning of the high-value target. The method is suitable for rapid identification and positioning of high-value targets under remote sensing images.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer algorithms, and in particular relates to a high-value target rapid detection and positioning algorithm based on ultra-wide sweep remote sensing images, so as to achieve rapid recognition and positioning of high-value targets in remote sensing images. Background Art

[0002] In the field of remote sensing, high-resolution imaging systems play a key role in obtaining detailed and accurate spatial information on the Earth's surface. Among these systems, the rotary scanning camera, as an important innovation, provides a unique advantage in capturing a wide ground coverage area with high precision. The swing scanning camera is designed to scan a wide area and capture continuous images using a rotating mechanism, which are then stitched together to form a continuous panoramic view of the target area. This imaging method can obtain regional information over a large area at one time, and the coverage width can easily reach thousands of kilometers, greatly shortening the revisit time and enhancing the ability to quickly perceive ships, aircraft and ground targets. However, the remote sensing images obtained by this imaging method face several key problems. First, due to the significant increase in coverage width, the image resolution is also significantly increased. How to input ultra-wide remote sensing images into the detector has become the primary problem of remote sensing image target detection. Second, due to the increase in the swing scanning angle, the distance between the object point and the image plane is also increasing, which leads to a sharp decrease in the ground sampling distance. The edge of the imaging area in the along-track direction is sub-pixelized, and the scenery on both sides of the image plane in the vertical direction will be compressed. In addition, in side-view scenarios (i.e., large sweep angles), the target object exhibits different feature information from the top-view scenario and causes greater geometric distortion, which makes remote sensing image target detection more difficult. How to overcome the above difficulties and achieve rapid detection and positioning of high-value targets in ultra-wide sweep remote sensing images is an important research topic due to its wide application in the real world.

[0003] With the rapid development of deep learning technology and the introduction of large-scale labeled remote sensing image datasets in recent years, remote sensing image target detection has achieved significant improvements in recognition accuracy and speed. However, how to achieve efficient target detection on ultra-wide remote sensing images still faces an important challenge: unlike general target detection tasks that directly input images into detectors, ultra-wide remote sensing images cannot be directly input into detectors due to the limitation of image processing unit (GPU) memory capacity. Moreover, if ultra-wide remote sensing images are downsampled to a size suitable for input detectors, information will be lost due to the reduction in resolution, especially in the detection of small-sized objects. At present, the paradigm proposed for this problem is: according to a fixed stride, the ultra-wide image is cropped to a size suitable for input detectors (such as 800×800 pixels) through a sliding window. After that, all slices are input into the detector and the detection data is mapped onto the original image for synthesis.

[0004] However, for ultra-wide-width swing-scan remote sensing images, due to the significant decrease in spatial resolution as the swing-scan angle increases, simply setting the overlap ratio based on experience and using the sliding window method to crop the image will result in two situations: when the overlap ratio is set too large, in places with large swing-scan angles, due to the low spatial resolution, the fixed-step segmentation will generate too many redundant slices, affecting the detection speed; when the overlap ratio is set too small, in places with small swing-scan angles, especially near the subsatellite point, due to the high spatial resolution, the fixed-step segmentation will cause the complete object to be segmented in two adjacent slices, resulting in misidentification, missed detection, and target positioning errors. In this case, a natural question is: Can we use a segmentation method that is more suitable for ultra-wide-width swing-scan remote sensing images to improve the detection speed while ensuring that the target detection and positioning accuracy is not reduced as much as possible? Summary of the invention

[0005] The purpose of the present invention is to provide a method for rapid detection and positioning of high-value targets based on ultra-wide-width rotary scanning remote sensing images, which is used to realize rapid identification and positioning of high-value targets in remote sensing images, while improving the detection speed while ensuring that the target detection and positioning accuracy does not decrease as much as possible.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The present invention provides a method for rapid detection and positioning of high-value targets based on ultra-wide sweep remote sensing images, the method comprising the following steps:

[0008] Step S1: Obtaining an ultra-wide swing-scan remote sensing image and the timestamp corresponding to each column of pixels in the image, remote sensing satellite orbit data, and swing-scan camera parameters;

[0009] Step S2: using the remote sensing satellite orbit data, the swing-scan camera parameters and the ultra-wide swing-scan remote sensing image, the overlap ratio required for the corresponding pixel position is calculated in real time, and the ultra-wide swing-scan remote sensing image is divided into slices of several sizes according to the overlap ratio;

[0010] Step S3: using slices of several sizes as input to the detector, outputting the target detection frame, and mapping the detection frame back to the ultra-wide remote sensing image according to the position of the slice in the original image;

[0011] Step S4: The pixel coordinates of the center of the target detection frame are used as the image position of the target, and the target pixel coordinates are mapped to the WGS84 coordinates to complete the detection and positioning of the high-value target;

[0012] Step S5: Use the test set, known target location information, and nearest neighbor matching to calculate the positioning error to measure the performance of the positioning method.

[0013] Furthermore, there is another preferred embodiment, the above step S1 is specifically as follows:

[0014] In a simulation environment, a scene is built for shooting the target with a swing-scan remote sensing camera, and the ultra-wide swing-scan remote sensing image after the detector is superimposed is output. At the same time, the timestamp corresponding to each column of pixels in the image, the remote sensing satellite orbit data and the swing-scan camera parameters are output.

[0015] Furthermore, in a preferred embodiment, the above-mentioned parameters of the swing-scan camera include detector resolution, pixel physical size, focal length and angle information.

[0016] Furthermore, there is a preferred embodiment that uses a spatial resolution model of a sweeping remote sensing satellite to adaptively segment the remote sensing image.

[0017] Furthermore, in another preferred embodiment, in the above step S3, a non-maximum suppression algorithm is used to suppress the overlapping targets.

[0018] Furthermore, there is a preferred embodiment that uses a rigorous geometric model of swing scanning imaging to map the target pixel coordinates to WGS84 coordinates.

[0019] Furthermore, there is another preferred embodiment, the above step S5 is specifically as follows:

[0020] The collected test set is used to test and compare the inference speed and mAP of the overall algorithm, and the known target position information and nearest neighbor matching are used to calculate the positioning error to measure the overall performance of the algorithm.

[0021] The method for rapid detection and positioning of high-value targets based on ultra-wide-width rotary scanning remote sensing images described in the present invention can be fully implemented by computer software. Therefore, correspondingly, the present invention also provides a system for rapid detection and positioning of high-value targets based on ultra-wide-width rotary scanning remote sensing images, the system comprising:

[0022] A storage device for obtaining ultra-wide swing-scan remote sensing images and the timestamp corresponding to each column of pixels in the image, remote sensing satellite orbit data, and swing-scan camera parameters;

[0023] A storage device for calculating the overlap ratio required for corresponding pixel positions in real time through remote sensing satellite orbit data, swing-scan camera parameters and ultra-wide swing-scan remote sensing images, and dividing the ultra-wide swing-scan remote sensing images into slices of several sizes according to the overlap ratio;

[0024] A storage device for taking slices of several sizes as inputs of a detector, outputting a target detection frame, and mapping the detection frame back to the ultra-wide remote sensing image according to the position of the slice in the original image;

[0025] A storage device for taking the pixel coordinates of the center of the target detection frame as the image position of the target and mapping the target pixel coordinates to WGS84 coordinates to complete the detection and positioning of high-value targets;

[0026] A storage device for measuring the performance of a positioning method using a test set, known target location information, and nearest neighbor matching to calculate positioning errors.

[0027] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for rapid detection and positioning of high-value targets based on ultra-wide-width rotary scanning remote sensing images described above is executed.

[0028] The present invention also provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes any one of the above-mentioned methods for rapid detection and positioning of high-value targets based on ultra-wide-width rotary scanning remote sensing images.

[0029] The beneficial effects of the present invention are:

[0030] 1. The present invention proposes a method for rapid detection and positioning of high-value targets based on ultra-wide-width rotary scanning remote sensing images, constructs a swing scanning remote sensing image data set, and realizes full-link simulation from swing scanning remote sensing images to high-precision longitude and latitude positions of targets. This method can effectively identify and locate high-value targets in ultra-wide-width rotary scanning remote sensing images. Compared with traditional methods, it achieves faster detection and positioning speed while ensuring that the accuracy does not decrease.

[0031] 2. The present invention solves the input problem of ultra-wide remote sensing images in traditional methods, realizes rapid detection and positioning of large images of any size, and ensures that the accuracy is not reduced. In addition, the present invention can be used in conjunction with the most advanced detectors to achieve faster detection speed and detection accuracy.

[0032] The present invention is used to realize rapid recognition and positioning of high-value targets in remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1It is a flow chart of the overall solution of a method for rapid detection and positioning of high-value targets based on ultra-wide sweep remote sensing images described in the present invention;

[0035] Figure 2 It is the characteristic diagram of the rotary scanning remote sensing image described in the present invention;

[0036] Figure 3 is a schematic diagram of the rotary sweep remote sensing of the present invention;

[0037] Figure 4 is a schematic diagram of the segmentation algorithm described in the present invention;

[0038] Figure 5 It is the remote sensing satellite orbit data of the present invention;

[0039] Figure 6 It is a relationship diagram between the optical center of the camera and the object point during the swing scanning imaging described in the present invention;

[0040] Figure 7 It is the ground pixel resolution model in the vertical track direction described in the present invention;

[0041] Figure 8 It is the ground pixel resolution model in the along-track direction described in the present invention;

[0042] Fig. 9 is an example diagram of the segmentation algorithm described in the present invention;

[0043] Fig.10 It is a strict geometric model of ultra-wide sweep imaging described in the present invention;

[0044] Fig.11 It is the time-mAP comparison diagram of the present invention;

[0045] Fig.12 It is a time-position error comparison diagram described in the present invention. DETAILED DESCRIPTION

[0046] The specific embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

[0047] Implementation method 1, see Figure 1 This embodiment provides a method for rapid detection and positioning of high-value targets based on ultra-wide sweep remote sensing images, which is used to achieve rapid recognition and positioning of high-value targets in remote sensing images. Specifically, the following steps are included: Figure 1 As shown:

[0048] Step 1: Build a scene in the simulation environment where the swing-scan remote sensing camera shoots the target, output the ultra-wide swing-scan remote sensing image after the detector is superimposed, and simultaneously output the timestamp, remote sensing satellite orbit data, and swing-scan camera parameters (including detector resolution, pixel physical size, focal length, and angle information) corresponding to each column of pixels in the image. The input of the overall system is satellite orbit data, camera parameters, and ultra-wide swing-scan remote sensing images;

[0049] Step 2: By using the input satellite orbit data, camera parameters and ultra-wide-width swing-scan remote sensing images, the overlap ratio required for the corresponding pixel position is calculated in real time, and the ultra-wide-width swing-scan remote sensing images are divided into several slices of appropriate size (for example, 800×800 pixels) as the input of the detector.

[0050] Step 3: Use several slices of the cropped image as the input of the detector, output the target detection frame, and map the detection frame back to the ultra-wide remote sensing image according to the position of the slice in the original image. For the problem of repeated detection of overlapping targets, the non-maximum suppression algorithm is used to suppress it.

[0051] Step 4: For the output detection frame information, take the pixel coordinates of the center of the detection frame as the image position of the target, and use the rigorous geometric model of the swing scanning imaging to map the target pixel coordinates to the WGS84 coordinates to achieve the detection and positioning of high-value targets.

[0052] Step 5: Use the collected test set to test and compare the inference speed and mAP of the overall algorithm. Use the known target location information and nearest neighbor matching to calculate the positioning error. This measures the overall performance of the algorithm.

[0053] The high-value target rapid detection and positioning method proposed in this embodiment realizes the full-link simulation from the swing-scan remote sensing image to the high-precision latitude and longitude position of the target by constructing a swing-scan remote sensing image dataset. This method can effectively identify and locate high-value targets in ultra-wide-width rotary-scan remote sensing images. Compared with traditional methods, it achieves faster detection and positioning speed while ensuring that the accuracy is not reduced.

[0054] Implementation Method 2: See Figures 2 to 12 This embodiment is described as follows: This embodiment specifically describes a method for rapid detection and positioning of high-value targets based on ultra-wide-width rotary scanning remote sensing images described in Embodiment 1;

[0055] Step 1: Build a scene in the simulation environment where the swing-scan remote sensing camera shoots the target, output the ultra-wide swing-scan remote sensing image after the detector is superimposed, and simultaneously output the timestamp, remote sensing satellite orbit data, and swing-scan camera parameters (including detector resolution, pixel physical size, focal length, and angle information) corresponding to each column of pixels in the image. The input of the overall system is satellite orbit data, camera parameters, and ultra-wide swing-scan remote sensing images;

[0056] Specifically include:

[0057] Simulation scene construction and data output, such as Figures 2 to 3 As shown in the figure, an ultra-wide sweep remote sensing satellite shooting scene is built in the Spacesim simulation system, and different types of ships and aircraft models are randomly placed in the scene as targets to be detected. Remote sensing satellites are used to shoot ground scenes to generate remote sensing images, and the orbit data of remote sensing satellites is output, such as Figure 5 shown.

[0058] Step 2: By using the input satellite orbit data, camera parameters and ultra-wide-width swing-scan remote sensing images, the overlap ratio required for the corresponding pixel position is calculated in real time, and the ultra-wide-width swing-scan remote sensing images are divided into several slices of appropriate size (for example, 800×800 pixels) as the input of the detector.

[0059] Specifically include:

[0060] Ultra-wide-width rotary scanning remote sensing image segmentation, since remote sensing images cannot be directly input into the detector for target detection, it is necessary to segment large-size remote sensing images into small slices suitable for input. Figure 4 As shown in FIG. 1 , the spatial resolution model of the sweep remote sensing satellite is used to realize the adaptive segmentation of large-size remote sensing images.

[0061] During the scanning period, the field of view angle corresponding to a single pixel of the area array CMOS camera is:

[0062]

[0063] like Figure 6 As shown in the figure, when the camera sweep angle is α, the angle η between OS and OD is obtained by the sine theorem:

[0064]

[0065] Among them, R is the radius of the earth and H is the altitude of the satellite orbit.

[0066] The distance between S and D (object distance) is:

[0067]

[0068] From this we can get, if Figure 7The ground pixel resolution in the vertical direction is shown:

[0069]

[0070] Among them, α is the field of view angle of the camera in the vertical direction, and x is the pixel coordinate in the vertical direction corresponding to the object point D.

[0071] Considering that Figure 8 The ground pixel resolution along the track direction is shown in Figure 1. The ground pixel resolution along the track direction is the length of the arc segment GG', that is:

[0072] l GG' =R(γ-γ') (5)

[0073] Among them, γ is the angle between OS and OG, and γ' is the angle between OS and OG'.

[0074] Approximately to the vertical track direction, using the sine theorem, γ and γ' can be obtained as:

[0075]

[0076] The ground pixel resolution can be obtained:

[0077]

[0078] Among them, θ is the field of view angle of the camera in the vertical direction, and y is the pixel coordinate in the vertical direction corresponding to the object point D.

[0079] For example, Fig. 9 As shown:

[0080] Given an original ultra-wide swing-scan remote sensing image, first set the absolute size of the overlap based on the prior information of the identified object (for example, determine the approximate size based on the type information of the identified object). Then, starting from the upper left corner of the original ultra-wide swing-scan remote sensing image (i.e., pixel coordinate (0,0)), take the initial 800x800 pixel patch along the track (pixel row direction) and vertical track (pixel column direction). Then, calculate the ground resolution corresponding to the edge pixels according to the swing-scan camera spatial resolution model, and calculate the image overlap ratio according to the preset overlap value to achieve image adaptive segmentation under variable resolution.

[0081] Step 3: Use several slices of the cropped image as the input of the detector, output the target detection frame, and map it back to the ultra-wide remote sensing image according to the position of the slice in the original image. For the problem of repeated detection of overlapping targets, the non-maximum suppression algorithm is used to suppress it.

[0082] Specifically include:

[0083] Target detection, using the image data output from step 1 above to establish a target detection dataset. This implementation method builds a scene of a target photographed by a swing-scan remote sensing camera in a simulation environment, with an orbit altitude of 450km, an orbit eccentricity of 0, and a swing angle range of 0° to 65° in the vertical direction of the track. In order to enhance the range covered by the dataset, different camera single pixel sizes and focal lengths are set in different swing angle ranges to simulate shooting conditions at different orbit altitudes. The initial dataset consists of 1,803 images, covering 22 object categories, including two major categories: aircraft and ships. The aircraft contains four subcategories, such as B52, B737, etc., and the ship contains 16 subcategories, such as Acasta, Aigle, etc. The image size is 800×800pixel. In order to improve the rotation invariance of the model, scaling, rotation, perspective transformation and other operations are performed on the dataset. The final dataset contains 28,848 images and 27,725 instances.

[0084] The detection model of this method is implemented using yolov5 as the framework. The ratio of training set to validation set is 7:3. The initial learning rate is set to 0.01. SGD is used as the optimizer for model training. The bitch used in the training process is 8, and there are 300 epochs in total. The confidence level during detection is 0.10, and the IOU threshold is 0.50. After detection, the detection boxes obtained by all patches are remapped to the original large image. In the overlapping area of ​​adjacent images, the non-maximum suppression algorithm (nms) is used to suppress the results, and the threshold is 0.1.

[0085] Step 4: For the output detection frame information, take the pixel coordinates of the center of the detection frame as the image position of the target, and use the rigorous geometric model of the swing scanning imaging to map the target pixel coordinates to the WGS84 coordinates to achieve the detection and positioning of high-value targets.

[0086] Specifically include:

[0087] Target position solution, after using the detector to obtain the pixel coordinates of the target in the image, it is necessary to map the pixel coordinates back to WGS84 coordinates to obtain the longitude and latitude information of the target.

[0088] Considering that Fig.10 The strict geometric imaging model of the sweep remote sensing satellite shown in the figure has:

[0089]

[0090] Generally, the imaging model is a series of transformations from the pixel coordinate system to the WGS84 coordinate system. For the array scanning satellite, a rotation matrix is ​​introduced Used to describe the geometric relationship between the equivalent area array CMOS and the camera. Rewritten as:

[0091]

[0092] Where [X,Y,Z] T is the coordinate of the ground point in the WGS84 coordinate system corresponding to the image point; GSD:6.2 is the WGS84 coordinate of the array scanning satellite when imaging in orbit; Represents the rotation matrix from the J2000 coordinate system to the WGS84 coordinate system; Represents the rotation matrix from the satellite orbit coordinate system to the J2000 coordinate system; Represents the rotation matrix from the satellite body coordinate system to the satellite orbit coordinate system; Represents the rotation matrix from the camera coordinate system to the satellite body coordinate system; Represents the rotation matrix from the equivalent array CMOS coordinate system to the camera coordinate system.

[0093] After establishing the corresponding relationship between each pixel of the image and the coordinates of the ground object point, the remote sensing image can be used to locate the ground target. Assuming that the elevation H of the ground target to be located is known, it can be considered that the target is located on the following ellipsoid:

[0094]

[0095] Among them, a is the major axis of the earth and b is the minor axis.

[0096] In summary:

[0097]

[0098] The two solutions of the equation represent the intersection of the vector corresponding to the image pixel coordinates and the earth ellipsoid. Substituting the root with the smaller |μ into the above formula (11) can obtain the coordinates of the object point D in the WGS-84 coordinate system.

[0099] The center of the detection frame is taken as the target position, the height is preset to 0.5 km, and the position calculation module is input to obtain the longitude and latitude corresponding to each point.

[0100] Step 5: Use the collected test set to test and compare the inference speed and mAP of the overall algorithm. Use the known target location information and nearest neighbor matching to calculate the positioning error. This measures the overall performance of the algorithm.

[0101] Specifically:

[0102] Fig.11 The figure shows a comparison of the inference speed and mAP of the overall algorithm. It can be seen from the figure that compared with the traditional algorithm, the AICM method proposed in this embodiment has achieved a higher inference speed while also achieving a certain improvement in mAP. Specifically, the inference speed has been improved by 7.39% and the mAP has been improved by 7.54%.

[0103] Fig.12 The figure shows a time-position error comparison diagram. It can be seen from the figure that compared with the traditional algorithm, the AICM method proposed in this embodiment has achieved a higher reasoning speed while also improving the positioning accuracy to a certain extent. Specifically, in the case of small sweep angles ([0,25]), the positioning accuracy is improved by 4.13%.

[0104] Implementation method 3: Any one of the above-mentioned methods for rapid detection and positioning of high-value targets based on ultra-wide-width rotary scanning remote sensing images can be implemented entirely by computer software. Therefore, correspondingly, this implementation method provides a system for rapid detection and positioning of high-value targets based on ultra-wide-width rotary scanning remote sensing images, and the system includes:

[0105] A storage device for obtaining ultra-wide swing-scan remote sensing images and the timestamp corresponding to each column of pixels in the image, remote sensing satellite orbit data, and swing-scan camera parameters;

[0106] A storage device for calculating the overlap ratio required for corresponding pixel positions in real time through remote sensing satellite orbit data, swing-scan camera parameters and ultra-wide swing-scan remote sensing images, and dividing the ultra-wide swing-scan remote sensing images into slices of several sizes according to the overlap ratio;

[0107] A storage device for taking slices of several sizes as inputs of a detector, outputting a target detection frame, and mapping the detection frame back to the ultra-wide remote sensing image according to the position of the slice in the original image;

[0108] A storage device for taking the pixel coordinates of the center of the target detection frame as the image position of the target and mapping the target pixel coordinates to WGS84 coordinates to complete the detection and positioning of high-value targets;

[0109] A storage device for measuring the performance of a positioning method using a test set, known target location information, and nearest neighbor matching to calculate positioning errors.

[0110] Implementation method 4: This implementation method provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, a method for rapid detection and positioning of high-value targets based on ultra-wide-angle sweep remote sensing images as described in any one of the above implementation methods is executed.

[0111] Embodiment 5. This embodiment provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes a method for rapid detection and positioning of high-value targets based on ultra-wide-width rotary scanning remote sensing images as described in any one of the above embodiments.

[0112] A computer device is provided in this embodiment. The hardware device of this part is a general model and is not shown in the form of a diagram. The system includes a processor and a memory, wherein the processor and the memory can be connected via a bus or other means. The memory is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules, and corresponding program instructions / modules. The processor executes various functional applications and data processing of the processor by running the non-volatile software programs, instructions and modules stored in the memory, so as to realize the method and steps for rapid detection and positioning of high-value targets based on ultra-wide-width rotary scanning remote sensing images in the above method embodiment.

[0113] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0114] The above description is only the implementation mode of the present invention and is not limited to the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of the claims of the present invention.

Claims

1. A method for rapid detection and positioning of high-value targets based on ultra-wide sweep remote sensing images, characterized in that: The method is: S1: Obtain ultra-wide swing-scan remote sensing images and the timestamp corresponding to each column of pixels in the image, remote sensing satellite orbit data, and swing-scan camera parameters; S2: The overlap ratio required for the corresponding pixel position is calculated in real time through remote sensing satellite orbit data, swing-scan camera parameters and ultra-wide swing-scan remote sensing images, and the ultra-wide swing-scan remote sensing images are divided into slices of several sizes according to the overlap ratio; S3: taking slices of several sizes as input to the detector, outputting the target detection box, and mapping the detection box back to the ultra-wide remote sensing image according to the position of the slice in the original image; S4: The pixel coordinates of the center of the target detection frame are used as the image position of the target, and the target pixel coordinates are mapped to the WGS84 coordinates to complete the detection and positioning of high-value targets; S5: The performance of the positioning method is measured using the test set, known target location information, and nearest neighbor matching to calculate the positioning error.

2. The method for rapid detection and positioning of high-value targets based on ultra-wide sweep remote sensing images according to claim 1 is characterized in that: S1 is specifically: In a simulation environment, a scene is built for shooting the target with a swing-scan remote sensing camera, and the ultra-wide swing-scan remote sensing image after the detector is superimposed is output. At the same time, the timestamp corresponding to each column of pixels in the image, the remote sensing satellite orbit data and the swing-scan camera parameters are output.

3. The method for rapid detection and positioning of high-value targets based on ultra-wide sweep remote sensing images according to claim 2 is characterized in that: The parameters of the swing-scan camera include detector resolution, pixel physical size, focal length, and angle information.

4. The method for rapid detection and positioning of high-value targets based on ultra-wide sweep remote sensing images according to claim 1 is characterized in that: The spatial resolution model of the sweep remote sensing satellite is used to adaptively segment the remote sensing images.

5. The method for rapid detection and positioning of high-value targets based on ultra-wide sweep remote sensing images according to claim 1 is characterized in that: In S3, non-maximum suppression algorithm is used to suppress overlapping targets.

6. The method for rapid detection and positioning of high-value targets based on ultra-wide sweep remote sensing images according to claim 1 is characterized in that: The strict geometric model of swing-scan imaging is used to map the target pixel coordinates to WGS84 coordinates.

7. The method for rapid detection and positioning of high-value targets based on ultra-wide sweep remote sensing images according to claim 1 is characterized in that: S5 is as follows: The collected test set is used to test and compare the inference speed and mAP of the overall algorithm, and the known target position information and nearest neighbor matching are used to calculate the positioning error to measure the overall performance of the algorithm.

8. A high-value target rapid detection and positioning system based on ultra-wide sweep remote sensing images, characterized in that: The system includes: A storage device for obtaining ultra-wide swing-scan remote sensing images and the timestamp corresponding to each column of pixels in the image, remote sensing satellite orbit data, and swing-scan camera parameters; A storage device for calculating the overlap ratio required for corresponding pixel positions in real time through remote sensing satellite orbit data, swing-scan camera parameters and ultra-wide swing-scan remote sensing images, and dividing the ultra-wide swing-scan remote sensing images into slices of several sizes according to the overlap ratio; A storage device for taking slices of several sizes as inputs of a detector, outputting a target detection frame, and mapping the detection frame back to the ultra-wide remote sensing image according to the position of the slice in the original image; A storage device for taking the pixel coordinates of the center of the target detection frame as the image position of the target and mapping the target pixel coordinates to WGS84 coordinates to complete the detection and positioning of high-value targets; A storage device for measuring the performance of a positioning method using a test set, known target location information, and nearest neighbor matching to calculate positioning errors.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, executes a method for rapid detection and positioning of high-value targets based on ultra-wide-width rotary scanning remote sensing images as described in any one of claims 1 to 7.

10. A computer device, characterized in that: The device includes a memory and a processor, wherein the memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes a method for rapid detection and positioning of high-value targets based on ultra-wide-width rotary scanning remote sensing images as described in any one of claims 1 to 7.

Images