A Microwave Remote Sensing Target Classification Optimization Method

Abstract

This disclosure provides an optimization method for microwave remote sensing target classification. In one specific embodiment, the method includes acquiring multiple target sample images of a region to be detected via a ground-based center; extracting the boundary point coordinates of the multiple target sample images and calculating the centroid coordinates of the multiple target sample images; constructing a target sample dataset; receiving and storing the target sample dataset via a satellite; observing the region to be detected, acquiring satellite observation intervals and observation images, reading the target sample dataset, and filtering out multiple target sample images located within the observation interval; classifying the observation images to obtain at least one classification result; calculating the matching degree between each classification result and the multiple target sample images located within the observation interval, and obtaining the matching result. This implementation further optimizes microwave remote sensing target classification under low memory and computing power conditions, improves classification accuracy, and solves the false alarm problem.

Description

Technical Field

[0001] This disclosure relates to the field of microwave remote sensing target classification. More specifically, it relates to an optimization method for microwave remote sensing target classification. Background Technology

[0002] Currently, to achieve ship route and port berthing planning and ensure safe navigation, a system capable of long-distance monitoring and classification, as well as global maritime surveillance, is needed. Constructing a remote sensing system that provides comprehensive monitoring of targets in various regions, with wide-area coverage capabilities, and effectively cataloging and managing sea surface targets, is crucial for maintaining safe navigation and has broad application value. This involves radiating electromagnetic signals from satellites into the scanned area, and acquiring target information by effectively detecting target echo signals through satellite receivers. In recent years, with the development of microwave remote sensing, the elimination of false alarms to improve target classification accuracy has become a hot topic of concern. However, due to the diverse marine environment, some sea surface elements, such as drilling platforms, lighthouses, and the surfaces of buildings on islands, have similar echo characteristics to ships. This makes it difficult for real-time sea surface target classification and processing technologies to distinguish between ships and sea surface elements with similar artificial surfaces, easily leading to false alarms and affecting target classification accuracy. To improve recognition accuracy, it is necessary to store target sample image data such as drilling platforms, lighthouses, and islands on satellite equipment, which needs to read and recognize these images during the classification process. However, the memory and computing power resources of satellite computing equipment are limited, while improving target classification accuracy requires a large amount of satellite equipment memory and computing power, making it impossible to solve the false alarm problem. Summary of the Invention

[0003] The purpose of this disclosure is to provide a microwave remote sensing target classification optimization method that occupies little satellite space, in order to solve at least one of the problems existing in the prior art.

[0004] To achieve the above objectives, the present disclosure adopts the following technical solution: The first aspect of this disclosure provides a microwave remote sensing target classification optimization method, including: Multiple target sample images within the area to be detected are obtained by acquiring images from the ground center. The coordinates of the boundary points of each target sample image are extracted, and the centroid coordinates of each target sample image are calculated. A target sample dataset is then constructed. The target sample dataset is acquired and stored via satellite; the area to be detected is observed to obtain the satellite observation range and observation images; the target sample dataset is read and multiple target sample images located within the observation range are selected; the observation images are classified to obtain at least one classification result; the matching degree of each classification result with the multiple target sample images located within the observation range is calculated and a matching result is obtained.

[0005] Furthermore, the constructed target sample dataset includes: Obtain the starting coordinates of the boundary points of the target sample image and calculate the difference between two adjacent boundary points in the boundary points of the target sample image to obtain the boundary point difference dataset; Construct the first target sample dataset and store the centroid coordinates and boundary point origin coordinates of multiple target sample images respectively; Construct a second target sample dataset and store the boundary point difference datasets of multiple target sample images respectively; Obtain the start and end positions of the boundary point difference dataset stored in the second target sample dataset for each target sample image, and store the start and end positions in the first target sample dataset respectively.

[0006] Furthermore, the method for calculating the start and end positions of the boundary point difference dataset stored in the second target sample dataset in each target sample image includes: Obtain the total number of differences between two adjacent coordinates in the boundary point coordinates of the i-th target sample image; According to the formula:

[0007]

[0008] Calculate the first The second sample data of the sample images is located at the starting position of the second target sample dataset. and end position .

[0009] Furthermore, obtaining the satellite observation range includes obtaining the coordinates of multiple vertices of the satellite observation range and calculating the four boundary points of the satellite observation range based on the coordinates of the multiple vertices.

[0010] Furthermore, the step of reading the target sample dataset and filtering out multiple target sample images located within the observation interval includes: Multiple first-filtered centroid coordinates are obtained by filtering the centroid coordinates located within the four cardinal points in the first target sample dataset; Starting from the first centroid coordinates of the first selection, output a first ray, calculate the intersection of the first ray with the boundary line of the observation interval, and determine whether the number of intersection points is even. If it is even, the first centroid coordinates of the first selection are located within the observation interval to obtain the second centroid coordinates of the second selection; otherwise, the first centroid coordinates of the first selection are located outside the observation interval. Read the first target sample dataset to obtain the starting coordinates, starting position, and ending position of the boundary point corresponding to the second selection centroid coordinates; read the second target sample dataset, obtain the corresponding boundary point difference dataset according to the starting position and ending position corresponding to the second selection centroid coordinates, calculate the corresponding boundary point coordinates, and fit multiple target sample images located in the observation interval.

[0011] Furthermore, the calculated coordinates of the corresponding boundary points include: According to the formula

[0012] The coordinates of the nth boundary point of the target sample image corresponding to the mth second screening centroid coordinates are calculated. ;in, Let these be the coordinates of the starting point of the boundary point of the target sample image corresponding to the centroid coordinates of the m-th second screening sample. It is the sum of the difference results between two adjacent boundary points in the coordinates of the first boundary point to the nth boundary point of the target sample image corresponding to the centroid coordinates of the mth second screening point.

[0013] Furthermore, the classification result is configured to be labeled on the probability circle of the observed image.

[0014] Further, calculating the matching degree between each of the classification results and the multiple target sample images located within the observation interval includes: Multiple target sample images located within the observation interval are acquired to obtain multiple target sample images to be matched; Perform the following operations on each classification result: Determine whether the probability circles of the multiple target sample images to be matched intersect with the classification results. If they intersect, calculate the first matching degree; otherwise, calculate the second matching degree. Each of the multiple target sample images to be matched is determined to have a first matching degree or a second matching degree greater than or equal to a set threshold. If it is greater than the threshold, the corresponding target sample image matches the classification result and the classification result is discarded. Otherwise, the corresponding target sample image does not match the classification result.

[0015] Further, the step of determining whether the probability circles of the plurality of target sample images to be matched intersect with the classification results, and calculating the first matching degree if they intersect, and the second matching degree otherwise, includes: Using the centroid of the plurality of target sample images to be matched as a reference, the plurality of target sample images to be matched are magnified by a length r to obtain a plurality of magnified target sample images; where r is the radius of the probability circle of the classification result; Starting from the center of the probability circle of the classification result, a second ray is output. The intersection points of the second ray with the boundaries of the magnified images of the multiple target samples are calculated. It is determined whether the number of intersection points is odd. If it is odd, the probability circle of the classification result intersects with the corresponding target sample image to be matched, and the first matching degree is calculated; otherwise, the probability circle of the classification result does not intersect with the corresponding target sample image to be matched, and the second matching degree is calculated. ;in The confidence level of the probability circle for the classification result.

[0016] Furthermore, the method for calculating the first matching degree includes: Perform the following operations on multiple target sample images to be matched: Multiple first distances are obtained by calculating the distances between the coordinates of multiple boundary points of the target sample image to be matched and the center of the probability circle of the classification result; multiple second distances are obtained by calculating the distances between the perpendicular lines of multiple boundary lines of multiple target sample images to be matched and the center of the probability circle of the classification result. The maximum distance is obtained by taking the maximum value among multiple first distances and multiple second distances. ; Obtain the minimum value among multiple second distances and multiple first distances respectively, and get the minimum distance. ;

[0017] The first matching degree is calculated. ,in, The radius of the probability circle for the classification results; The confidence level of the probability circle for the classification results; The area of ​​the intersection region between the probability circle of the classification result and the image of the target sample to be matched; The origin coordinates are used to place the fixed-type target at the origin coordinates in the coordinate axis. This represents the total number of boundary points in the target sample image to be matched; and These are the coordinates of the (i-1)th boundary point and the ith boundary point on the target sample image to be matched, respectively.

[0018] The beneficial effects of this disclosure are as follows: This invention pre-extracts the boundary point coordinates and centroid coordinates of target sample images such as drilling platforms, lighthouses, and islands from the ground center and constructs a target sample dataset to replace the original sample images. This reduces the amount of data stored and processed on the satellite by an order of magnitude, perfectly adapting to scenarios with limited satellite memory and computing power. Furthermore, by calculating the matching degree between the classification results and the centroid and boundaries of the target samples, this invention can effectively distinguish ships from static sample targets such as drilling platforms, lighthouses, and islands with similar echo characteristics. This further optimizes microwave remote sensing target classification under low memory and computing power environments, improving classification accuracy and effectively solving the false alarm problem. Attached Figure Description

[0019] The specific embodiments of this disclosure will be described in further detail below with reference to the accompanying drawings.

[0020] Figure 1 The flowchart of the microwave remote sensing target classification optimization method of this disclosure is shown.

[0021] Figure 2 A schematic diagram of the structure of a computer system for implementing the apparatus provided in the embodiments of this disclosure is shown. Detailed Implementation

[0022] To more clearly illustrate this disclosure, the following description, in conjunction with embodiments and accompanying drawings, provides further insight. Similar components in the drawings are indicated by the same reference numerals. Those skilled in the art should understand that the specific description below is illustrative rather than restrictive and should not be construed as limiting the scope of protection of this disclosure.

[0023] like Figure 1 As shown, one embodiment of this disclosure provides a microwave remote sensing target classification optimization method, including: Step S10: Obtain multiple target sample images within the area to be detected through the ground center, extract the boundary point coordinates of the multiple target sample images respectively, and calculate the centroid coordinates of the multiple target sample images respectively; construct the target sample dataset; Step S20: Acquire and store the target sample dataset via satellite; observe the area to be detected, acquire the satellite observation range and observation images, read the target sample dataset and filter out multiple target sample images located in the observation range; classify the observation images to obtain at least one classification result, calculate the matching degree between each classification result and the multiple target sample images located in the observation range and obtain the matching result.

[0024] This invention pre-extracts the boundary point coordinates and centroid coordinates of target sample images such as drilling platforms, lighthouses, and islands from the ground center and constructs a target sample dataset to replace the original sample images. This reduces the amount of data stored and processed on the satellite by an order of magnitude, perfectly adapting to scenarios with limited satellite memory and computing power. Furthermore, by calculating the matching degree between the classification results and the centroid and boundaries of the target samples, this invention can effectively distinguish ships from static sample targets such as drilling platforms, lighthouses, and islands with similar echo characteristics. This further optimizes microwave remote sensing target classification under low memory and computing power environments, improving classification accuracy and effectively solving the false alarm problem.

[0025] In a specific example, the satellite observes the acquired observation range and images using microwave remote sensing technology. Specifically, the observation images are microwave remote sensing images. In the microwave remote sensing images of this invention, the effective identification features of man-made facilities such as drilling platforms and lighthouses, as well as fixed targets such as islands, are mainly determined by their geometric boundaries, i.e., the contours formed by strong scattering points, rather than texture or color information. This allows for effective characterization of target samples through boundary point coordinates. Secondly, satellite platforms have extremely limited memory and computing resources, while massive amounts of prior sample data need to be stored to achieve global maritime surveillance. This requires that the sample data be compressed and stored in a way with extremely low redundancy, while ensuring rapid recovery and matching during real-time processing. Thirdly, fixed targets have the prior characteristic of basically unchanged geographical coordinates, while the observation results change dynamically. Therefore, the geometric matching of dynamic observation results with static prior samples in this invention is feasible. Fourthly, microwave remote sensing classification results inherently contain positioning errors and confidence information, which are usually output in the form of probability circles. In summary, the microwave remote sensing target classification optimization method proposed in this invention extracts and differentially encodes the boundary points and centroids of target sample images from the ground center to construct a lightweight dataset. This dataset is stored on the satellite and used for rapid screening, fitting, and matching degree calculation during observation, thereby achieving high-precision classification in a low-memory and low-computing-power environment and effectively eliminating false alarms.

[0026] In a specific example, after the ground center constructs the target sample dataset, it sends the target sample dataset to the satellite; the satellite then obtains and stores the target sample dataset sent by the ground center.

[0027] In another example, after the ground center constructs the target sample dataset, before the satellite launch, the target sample dataset can be copied to a computer-readable medium on the satellite so that the satellite can pre-store the target sample dataset and use it for the execution of step S20.

[0028] In one possible implementation, the construction of the target sample dataset includes: Obtain the starting coordinates of the boundary points of the target sample image and calculate the difference between two adjacent boundary points in the boundary points of the target sample image to obtain the boundary point difference dataset; Construct the first target sample dataset and store the centroid coordinates and boundary point origin coordinates of multiple target sample images respectively; Construct a second target sample dataset and store the boundary point difference datasets of multiple target sample images respectively; Obtain the start and end positions of the boundary point difference dataset stored in the second target sample dataset for each target sample image, and store the start and end positions in the first target sample dataset respectively.

[0029] In a specific example, the first target sample dataset can be a first data table; the second target sample dataset can be a second data table. In this embodiment, for ease of storage, the data needs to be simplified. Key points can be retained using the Douglas-Pokal algorithm, and the maximum allowable vertical distance between each vertex and the newly created line can be set to calculate the simplified planar data. The fitting points and centroids of each planar data are extracted, and two tables are used to store the coordinates of the fitting points and centroids of each planar data. To reduce the required storage space, the first data table stores the starting coordinates (i.e., starting latitude and longitude), centroid coordinates (i.e., centroid latitude and longitude), and the difference results between subsequent points and previous points (the first previous point is the starting point) in the boundary points (these results are multiplied by a constant to obtain integer data) in the starting and ending positions of the second data table. The second data table stores the difference results of the latitude and longitude between subsequent points and previous points in column form.

[0030] In a specific example, the target sample image is processed by the ground center into planar vector data so that the ground center can extract the coordinates of the boundary points of multiple target sample images respectively; Multiple target sample images are defined as follows: , with target sample image For example, the set of its boundary points is defined as Target sample image The centroid coordinates are The first target sample dataset stores , and target sample images The starting and ending positions of the boundary point difference dataset stored in the second target sample dataset are... The starting position of the sample dataset. =1, End position It is m-1.

[0031] The second target sample dataset stores the target sample images. The boundary point difference dataset; respectively .

[0032] In one possible implementation, the method for calculating the start and end positions of the boundary point difference dataset stored in the second target sample dataset in each target sample image includes: Get the i The total number of differences between adjacent coordinates in the boundary point coordinates of a target sample image. ; According to the formula:

[0033]

[0034] Calculation yields the first The second sample data of the sample images is located at the starting position of the second target sample dataset. and end position .

[0035] In one possible implementation, obtaining the satellite observation range includes obtaining the coordinates of multiple vertices of the satellite observation range and calculating the four boundary points of the satellite observation range based on the coordinates of the multiple vertices.

[0036] In a specific example, the satellite observation area is a quadrilateral with four vertices: , , as well as According to the formula:

[0037]

[0038]

[0039]

[0040] The four boundary points of the obtained satellite observation range are respectively , , as well as .

[0041] In one possible implementation, reading the target sample dataset and filtering out multiple target sample images located within the observation interval includes: Multiple first-filtered centroid coordinates are obtained by filtering the centroid coordinates located within the four cardinal points in the first target sample dataset; Starting from the first centroid coordinates of the first selection, output a first ray, calculate the intersection of the first ray with the boundary line of the observation interval, and determine whether the number of intersection points is even. If it is even, the first centroid coordinates of the first selection are located within the observation interval to obtain the second centroid coordinates of the second selection; otherwise, the first centroid coordinates of the first selection are located outside the observation interval. Read the first target sample dataset to obtain the starting coordinates, starting position, and ending position of the boundary points corresponding to the second selection centroid coordinates; read the second target sample dataset, obtain the corresponding boundary point difference dataset based on the starting and ending positions corresponding to the second selection centroid coordinates, calculate the corresponding boundary point coordinates, and fit multiple target sample images located in the observation interval. Specifically, in this embodiment, fitting refers to obtaining the corresponding target sample images located in the observation interval from the coordinates of multiple boundary points corresponding to the second selection centroid coordinates.

[0042] In a specific example, to filter target sample images whose centroid coordinates lie within the four boundary points, the following formula must be satisfied:

[0043] in, Let be the centroid coordinates of the i-th target sample image. The first selection centroid coordinates are obtained from these. The first ray is output using the first selected centroid coordinates as the starting point; it should be noted that the direction of the first ray can be any direction in the plane, and the observation interval can be calculated separately according to the two-point formula. The boundary line formed by two adjacent boundary points is obtained, and the intersection points of the first ray and multiple boundary lines are obtained. It is determined whether the number of intersection points is even. If it is even, the first selection centroid coordinates are located within the observation interval to obtain the second selection centroid coordinates. Otherwise, the first selection centroid coordinates are located outside the observation interval.

[0044] In one possible implementation, the calculation of the corresponding boundary point coordinates is equivalent to decompressing multiple target sample images located within the observation interval from the target sample dataset. The specific process includes: According to the formula

[0045] The coordinates of the nth boundary point of the target sample image corresponding to the mth second screening centroid coordinates are calculated. ;in, Let these be the coordinates of the starting point of the boundary point of the target sample image corresponding to the centroid coordinates of the m-th second screening sample. It is the sum of the difference results between two adjacent boundary points in the coordinates of the first boundary point to the nth boundary point of the target sample image corresponding to the centroid coordinates of the mth second screening point.

[0046] In one possible implementation, the classification result is configured to be calibrated to a probability circle on the observed image.

[0047] In a specific example, the microwave remote sensing classification result has a probability circle with radius r and a confidence level of [value missing]. In microwave remote sensing target classification, the probability circle represents the classification result and its uncertainty: the latitude and longitude coordinates of its center indicate the most likely location of the target as determined by the system; the length of its radius defines the range of the circular area centered on the center where the true location of the target may exist, which incorporates the system's positioning error; and the confidence score is a statistical probability value that indicates the likelihood of the target's true location falling within the circle defined by this radius, such as 68.27% or 99.73%. Together, these three factors constitute a complete and measurable spatial confidence interval.

[0048] In one possible implementation, calculating the matching degree between each of the classification results and the plurality of target sample images located within the observation interval includes: Multiple target sample images located within the observation interval are acquired to obtain multiple target sample images to be matched; Perform the following operations on each classification result: Determine whether the probability circles of the multiple target sample images to be matched intersect with the classification results. If they intersect, calculate the first matching degree; otherwise, calculate the second matching degree. Each of the multiple target sample images to be matched is determined to have a first matching degree or a second matching degree greater than or equal to a set threshold. If it is greater than the threshold, the corresponding target sample image matches the classification result and the classification result is discarded. Otherwise, the corresponding target sample image does not match the classification result.

[0049] In one possible implementation, determining whether the probability circles of the plurality of target sample images to be matched intersect with the classification results, and calculating the first matching degree if they intersect, and the second matching degree otherwise, includes: Using the centroid of the plurality of target sample images to be matched as a reference, the plurality of target sample images to be matched are magnified by a length r to obtain a plurality of magnified target sample images; where r is the radius of the probability circle of the classification result; Starting from the center of the probability circle of the classification result, a second ray is output. The intersection points of the second ray with the boundaries of the magnified images of the multiple target samples are calculated. It is determined whether the number of intersection points is odd. If it is odd, the probability circle of the classification result intersects with the corresponding target sample image to be matched, and the first matching degree is calculated; otherwise, the probability circle of the classification result does not intersect with the corresponding target sample image to be matched, and the second matching degree is calculated. ;in The confidence level of the probability circle representing the classification result. It should be noted that the direction of the second ray can be any direction; this embodiment does not impose any restrictions on this.

[0050] In a specific example, when determining whether the microwave remote sensing classification result intersects with the polygon of the target sample image, directly performing geometric intersection calculations between the circle and the polygon would be complex and computationally intensive, making it difficult to meet the resource constraints of real-time processing on the satellite. Therefore, this invention employs an efficient geometric equivalence transformation method, which involves enlarging the polygon of the target sample image outward by a distance *r* from its centroid to obtain the polygon of the enlarged target sample image. According to geometric principles, a probability circle intersects the original polygon if and only if the center of the probability circle lies inside the polygon of the enlarged target sample image. Thus, the complex circle-polygon intersection judgment is transformed into a simple point-polygon inclusion relationship judgment. After the transformation, only a ray needs to be drawn from the center of the circle, and the judgment can be completed by calculating the number of intersection points between the ray and the boundary of the enlarged target sample image. This transformation significantly reduces the computational complexity of the algorithm, enabling the satellite to complete false target matching and removal in real-time and accurately with limited memory and computing power.

[0051] In one possible implementation, the method for calculating the first matching degree includes: Perform the following operations on multiple target sample images to be matched: Multiple first distances are obtained by calculating the distances between the coordinates of multiple boundary points of the target sample image to be matched and the center of the probability circle of the classification result; multiple second distances are obtained by calculating the distances between the perpendicular lines of multiple boundary lines of multiple target sample images to be matched and the center of the probability circle of the classification result. The maximum distance is obtained by taking the maximum value among multiple first distances and multiple second distances. ; Obtain the minimum value among multiple second distances and multiple first distances respectively, and get the minimum distance. ;

[0052] The first matching degree is calculated. ,in, The radius of the probability circle for the classification results; The confidence level of the probability circle for the classification results; The area of ​​the intersection region between the probability circle of the classification result and the image of the target sample to be matched; The origin coordinates are used to place the fixed-type target at the origin coordinates in the coordinate axis. This represents the total number of boundary points in the target sample image to be matched; and These are the coordinates of the (i-1)th boundary point and the ith boundary point on the target sample image to be matched, respectively.

[0053] In a specific example, different target sample images to be matched each have their corresponding origin coordinates. In this embodiment, the origin coordinates Located within the target sample image to be matched, specifically, the centroid coordinates of the target sample image can be used as the origin coordinates. .

[0054] In a specific example, referring to the formula above, if If the target sample image to be matched is located inside the probability circle, then... If the probability circle lies within the image of the target sample to be matched; if If the probability circle partially overlaps with the image of the target sample to be matched, then the probability circle will be in a certain position.

[0055] In a specific example, the calculation process of calculating the matching degree between each classification result and multiple target sample images located in the observation interval, and obtaining the matching result, in this invention includes: In a specific example, this embodiment takes the target sample image as an island and expands the range of each island. Analyze whether the center of the probability circle is located within the expanded island. If not, continue analyzing the next island. If it is located within the island, proceed to the next step of analysis. 2) If the distance from the center of the circle to each point on the island is ≤ If the island is located inside the probability circle of the classification result, the matching degree is obtained by multiplying the ratio of the island's area to the circle's area by the confidence level. If > Then proceed to the next step of analysis; 3) Determine the shortest distance from the center of the circle to the polygon. If the shortest distance is greater than 1 / 3, then... If the classification result is located inside the island, then the matching degree is 100% × confidence level. If the shortest distance ≤ Then, calculate the ratio of the area of ​​the intersection between the classification result and the island to the area of ​​the probability circle of the classification result, and multiply it by the confidence level to obtain the matching degree between the two. 4) Obtain its threshold through ground training. If the matching degree is greater than This indicates that the classification result matches the island; if the classification is incorrect, it will be removed. The classification result is a probability circle in form, and can be a classification of ships in content.

[0056] In a specific example, the execution process of the method of the present invention includes, for instance, setting a target classification optimization algorithm on a certain project.

[0057] Based on the size of the ships of interest, global island vector data was filtered, resulting in more than 100,000 data points. The original required storage space was 270MB. Following the above process, a potential target sample database was formed, reducing the storage space to 10MB, which meets the storage requirements.

[0058] A total of over 100,000 data entries from islands were screened, and a potential target sample database was formed based on the above method. This database requires approximately 10MB of storage space to meet the onboard storage requirements.

[0059] Based on the range of a single satellite observation, this range is expanded by 10km. Next, a subset of target sample data is selected from the resulting potential target sample database. 1) Calculate the four boundary points of the satellite observation range, and filter the data of island targets whose centroids are located within the four boundary points. If they are located, proceed to 2). If not, continue to filter the next island target.

[0060] 2) Filter the data whose centroid is located on the edge of the satellite observation interval. If it is, perform restoration. The second point is obtained by adding the starting coordinates to the first column of difference results, the third point is obtained by adding the starting coordinates to the first and second columns of difference results, and so on, to obtain the coordinates of each fitted point of the restored island target; store it in the cache. Otherwise, continue to step 3).

[0061] 3) Use the ray method to screen island targets whose centroid is located within the satellite observation range. That is, draw a ray to the right from the centroid. If the number of intersections between the ray and the satellite observation range is odd, then restore the island target and store it in the cache. If the number of intersections is even, then start from 1) to continue screening the next island target.

[0062] Approximately 400 island fitting points were obtained, and the classification result is known to be a probability circle. , radius is Confidence level is That is, the focus result is located in The probability of being inside the circle with a radius of 1km(1σ) is 68.27%. The probability of finding the target within a circle with a radius of 3km (3σ) is 99.73%. This result can be considered as a circle with a radius of 3km. The geometric relationship between this 3km radius circle and the targets on each island is then analyzed.

[0063] Expand the area of ​​each island by 3km and analyze whether the center of the circle is located within the expanded island. If not, continue to analyze the next island; if it is, proceed to the next step of analysis. If the distance from the center of the circle to each point on the island is ≤3km, then the island is located inside the probability circle of the classification result. The matching degree between the two is obtained by calculating the product of the ratio of the island area to the circle area and the confidence level. If it is >3km, then proceed to the next step of analysis. Determine the shortest distance from the center of the circle to the polygon. If the shortest distance is greater than 3km, it indicates that the classification result is located inside the island, and the matching degree is 100% × confidence level. That is, 99.73%. If the shortest distance is ≤3km, the area of ​​the intersection between the classification result and the island is calculated as the ratio of the area of ​​the probability circle of the classification result to the area of ​​the intersection. The result is then multiplied by the confidence level to obtain the matching degree between the two. Its threshold is obtained through ground training. If the matching degree is greater than If so, then it will be removed.

[0064] In summary, firstly, this invention forms a potential target sample database based on geographic data and uses a method of matching microwave remote sensing classification results with samples to eliminate fixed-class targets with similar characteristics to ships, which can effectively reduce false alarms, improve classification accuracy, and optimize microwave remote sensing target classification results. Secondly, by matching and analyzing microwave remote sensing classification results with the stored database, this invention reduces the complexity of algorithm processing and the amount of computation required, which is beneficial for real-time processing in engineering practice and ensures the real-time performance of the entire process. Finally, by processing geographic data, this invention can reduce the storage space required for geographic data and meet storage needs.

[0065] like Figure 2 As shown, a computer system suitable for implementing the ground center or satellite provided in the above embodiments includes a central processing module (CPU), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) or a program loaded from a storage portion into a random access memory (RAM). The RAM also stores various programs and data required for the operation of the computer system. The CPU, ROM, and RAM are connected via a bus. An input / output (I / O) interface is also connected to the bus.

[0066] The following components are connected to the I / O interface: input sections including keyboards, mice, etc.; output sections including liquid crystal displays (LCDs) and speakers, etc.; storage sections including hard disks, etc.; and communication sections including network interface cards such as LAN cards and modems. The communication sections perform communication processing via networks such as the Internet. Drives are also connected to the I / O interface as needed. Removable media, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on the drive as needed so that computer programs read from them can be installed into the storage section as required.

[0067] Specifically, according to this embodiment, the process described in the flowchart above can be implemented as a computer software program. For example, this embodiment includes a computer program product comprising a computer program tangibly embodied on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium.

[0068] The flowcharts and schematic diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of the system, method, and computer program product of this embodiment. In this regard, each block in the flowchart or schematic diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the schematic diagram and / or flowchart, and combinations of blocks in the schematic diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0069] In the description of this disclosure, it should be noted that the terms "upper," "lower," etc., indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this disclosure and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this disclosure. Unless otherwise expressly specified and limited, the terms "installed," "connected," and "linked" should be interpreted broadly, for example, they can be fixed connections, detachable connections, or integral connections; they can be mechanical connections or electrical connections; they can be direct connections or indirect connections through an intermediate medium; they can be internal connections between two elements. For those skilled in the art, the specific meaning of the above terms in this disclosure can be understood according to the specific circumstances.

[0070] It should also be noted that, in the description of this disclosure, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0071] Obviously, the above embodiments of this disclosure are merely examples for clearly illustrating this disclosure, and are not intended to limit the implementation of this disclosure. For those skilled in the art, other variations or modifications can be made based on the above description. It is impossible to exhaustively list all implementation methods here. Any obvious variations or modifications derived from the technical solutions of this disclosure are still within the protection scope of this disclosure.

Claims

1. A microwave remote sensing target classification optimization method, characterized in that, include: Multiple target sample images within the area to be detected are obtained by acquiring images from the ground center. The coordinates of the boundary points of each target sample image are extracted, and the centroid coordinates of each target sample image are calculated. A target sample dataset is then constructed. The target sample dataset is acquired and stored via satellite. The area to be detected is observed to obtain satellite observation range and observation images. The target sample dataset is read and multiple target sample images located in the observation range are selected. The observation images are classified to obtain at least one classification result. The matching degree of each classification result with the multiple target sample images located in the observation range is calculated to obtain the matching result.

2. The method according to claim 1, characterized in that, The constructed target sample dataset includes: Obtain the starting coordinates of the boundary points of the target sample image and calculate the difference between two adjacent boundary points in the boundary points of the target sample image to obtain the boundary point difference dataset; Construct the first target sample dataset and store the centroid coordinates and boundary point origin coordinates of multiple target sample images respectively; Construct a second target sample dataset and store the boundary point difference datasets of multiple target sample images respectively; Obtain the start and end positions of the boundary point difference dataset stored in the second target sample dataset for each target sample image, and store the start and end positions in the first target sample dataset respectively.

3. The method according to claim 2, characterized in that, The method for calculating the start and end positions of the boundary point difference dataset stored in the second target sample dataset in each target sample image includes: Get the i The total number of differences between adjacent coordinates in the boundary point coordinates of a target sample image. ; According to the formula: Calculation yields the first The second sample data of the sample images is located at the starting position of the second target sample dataset. and end position .

4. The method according to claim 2, characterized in that, The process of obtaining the satellite observation range includes obtaining the coordinates of multiple vertices of the satellite observation range and calculating the four boundary points of the satellite observation range based on the coordinates of the multiple vertices.

5. The method according to claim 4, characterized in that, The step of reading the target sample dataset and filtering out multiple target sample images located within the observation interval includes: Multiple first-filtered centroid coordinates are obtained by filtering the centroid coordinates located within the four cardinal points in the first target sample dataset; Starting from the first centroid coordinates of the first selection, output the first ray, calculate the intersection of the first ray with the boundary line of the observation interval, and determine whether the number of intersection points is even. If it is even, the first centroid coordinates of the first selection are located within the observation interval to obtain the second centroid coordinates of the second selection; otherwise, the first centroid coordinates of the first selection are located outside the observation interval. Read the first target sample dataset to obtain the starting coordinates, starting position, and ending position of the boundary point corresponding to the second selection centroid coordinates; read the second target sample dataset, obtain the corresponding boundary point difference dataset according to the starting position and ending position corresponding to the second selection centroid coordinates, calculate the corresponding boundary point coordinates, and fit multiple target sample images located in the observation interval.

6. The method according to claim 5, characterized in that, The calculated coordinates of the corresponding boundary points include: According to the formula The coordinates of the nth boundary point of the target sample image corresponding to the mth second screening centroid coordinates are calculated. ;in, Let these be the coordinates of the starting point of the boundary point of the target sample image corresponding to the centroid coordinates of the m-th second screening sample. It is the sum of the difference results between two adjacent boundary points in the coordinates of the first boundary point to the nth boundary point of the target sample image corresponding to the centroid coordinates of the mth second screening point.

7. The method according to claim 1, characterized in that, The classification result is configured as a probability circle labeled on the observed image.

8. The method according to claim 7, characterized in that, The calculation of the matching degree between each classification result and multiple target sample images located in the observation interval includes: Multiple target sample images located within the observation interval are acquired to obtain multiple target sample images to be matched; Perform the following operations on each classification result: Determine whether the probability circles of the multiple target sample images to be matched intersect with the classification results. If they intersect, calculate the first matching degree; otherwise, calculate the second matching degree. Each of the multiple target sample images to be matched is determined to have a first matching degree or a second matching degree greater than or equal to a set threshold. If it is greater than the threshold, the corresponding target sample image matches the classification result and the classification result is discarded. Otherwise, the corresponding target sample image does not match the classification result.

9. The method according to claim 8, characterized in that, The process involves determining whether the probability circles of the multiple target sample images to be matched intersect with the classification results; if they intersect, the first matching degree is calculated. Otherwise, the calculated second matching degree includes: Using the centroid of the plurality of target sample images to be matched as a reference, the plurality of target sample images to be matched are magnified by a length r to obtain a plurality of magnified target sample images; where r is the radius of the probability circle of the classification result; Starting from the center of the probability circle of the classification result, a second ray is output. The intersection points of the second ray with the boundaries of the magnified images of the multiple target samples are calculated. It is determined whether the number of intersection points is odd. If it is odd, the probability circle of the classification result intersects with the corresponding target sample image to be matched, and the first matching degree is calculated; otherwise, the probability circle of the classification result does not intersect with the corresponding target sample image to be matched, and the second matching degree is calculated. ;in The confidence level of the probability circle for the classification result.

10. The method according to claim 9, characterized in that, The method for calculating the first matching degree includes: Perform the following operations on multiple target sample images to be matched: Multiple first distances are obtained by calculating the distances between the coordinates of multiple boundary points of the target sample image to be matched and the center of the probability circle of the classification result; multiple second distances are obtained by calculating the distances between the perpendicular lines of multiple boundary lines of multiple target sample images to be matched and the center of the probability circle of the classification result. The maximum distance is obtained by taking the maximum value among multiple first distances and multiple second distances. ; Obtain the minimum value among multiple second distances and multiple first distances respectively, and get the minimum distance. ; The first matching degree is calculated. ,in, The radius of the probability circle for the classification results; The confidence level of the probability circle for the classification results; The area of ​​the intersection region between the probability circle of the classification result and the image of the target sample to be matched; The origin coordinates are used to place the fixed-type target at the origin coordinates in the coordinate axis. This represents the total number of boundary points in the target sample image to be matched; and These are the coordinates of the (i-1)th boundary point and the ith boundary point on the target sample image to be matched, respectively.

Images