Ship target detection method of single-band image, electronic device and storage medium

Abstract

This invention relates to a method, electronic device, and storage medium for ship target detection using single-band imagery. The ship target detection method includes the following steps: performing rapid sliding window detection using single-band imagery and a pre-processed mask as input; outputting a single-frame target detection result after false alarm rejection; filtering for real ship targets based on the motion characteristics of the ship targets and the single-frame target detection result; and generating the position information and motion trajectory of the ship targets. This invention facilitates rapid detection of maritime ship targets and has high time and detection performance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of marine ship target detection, and in particular to a ship target detection method based on single-band image, an electronic device and a storage medium. BACKGROUND

[0002] The imaging information of ship targets in high-orbit single-band sequence images is limited, and the available elements are few. Even large ships at sea often exhibit target characteristics with only a few dozen or even a few pixels, and lack texture information, which cannot be distinguished by surface fine features. The currently rapidly developed machine learning target detection method through model training has the ability to detect small ship targets to some extent, but the algorithm volume is large, and when multiple images are detected and analyzed, more information needs to be processed, and GPU support is required, which has high requirements for hardware. In view of the imaging characteristics of the current high-orbit single-band remote sensing image, the ship target detection algorithm has the following problems:

[0003] (1) The ship target at sea is small and weak, the contrast between the target and the background gray value in the image is low, and the number of target pixels is very small. The interference of cloud layer is strong, and many clouds similar to the target often detect false targets, resulting in high false alarm rate.

[0004] (2) The single-band imaging data of high-orbit has only one spectrum, and there is no other observation data to assist detection, so it is impossible to use the multi-band linkage detection method in the form of decision tree to eliminate false alarms. Therefore, it is difficult to extract ship targets using limited observation information. If a "one-size-fits-all" detection mode is used, high false alarm rate or low detection rate will often occur.

[0005] (3) The detection method of deep learning depends on a large number of data sample training sets as input, and the algorithm portability is difficult. In addition, it needs to be supported and adapted by GPU and other hardware environments, and the calculation process is complex and slow, which is difficult to meet the real-time detection performance of multiple images. SUMMARY

[0006] In view of the above technical problems, the present application provides a ship target detection method based on single-band image, electronic device and storage medium, which can not rely on training samples and fully utilize single-band image for target detection, solving the problems of high false alarm rate in single-band image target detection and long traversal time in detection process.

[0007] The technical solution for achieving the purpose of the present application is as follows: a ship target detection method based on single-band image, comprising the following steps:

[0008] Step S1, using single-band image and preprocessed mask as input for fast sliding window detection, and outputting single-frame target detection result after false alarm elimination;

[0009] Step S2, according to the motion characteristics of the ship target and the single-frame target detection result, a real ship target is screened;

[0010] Step S3, position information and a motion trajectory of the ship target are generated.

[0011] According to one aspect of the present application, in the step S1, specifically comprising:

[0012] Step S11, taking high-orbit satellite single-band image data as input, auxiliary information and imaging information of the single-band image are obtained;

[0013] Step S12, double-parameter constant false alarm rate detection processing is performed on each pixel of the original single-band image, the to-be-detected pixel is taken as a target window, a rectangular area of 10 pixels extended from the center of the target window is taken as a protection window, and a ring area of 10 pixels to 20 pixels extended from the center of the target window is taken as a clutter window;

[0014] Step S13, DN value characteristics of the original image of the clutter window are counted to determine a segmentation threshold of the target window detection;

[0015] Step S14, adaptive threshold pixel classification is completed by using the segmentation threshold to obtain a to-be-determined target pixel set;

[0016] Step S15, after four rounds of false alarm elimination of the to-be-determined target pixel set, single-frame target detection data are obtained.

[0017] According to one aspect of the present application, in the step S13, the DN value characteristics of the original image of the clutter window are counted, specifically comprising:

[0018] It is assumed that the clutter background observation value obeys a normal distribution, and the probability density function is:

[0019]

[0020] It is assumed that the detection threshold is x=X, and the false alarm probability is p, and the following formula is established:

[0021]

[0022] When x=5, F(5)=0.99999, the detection threshold X is obtained as:

[0023] X=σ·Φ -1 (1-p)+μ

[0024] The above formula is taken as the segmentation threshold of the target window detection.

[0025] According to one aspect of the present application, in the step S15, specifically comprising:

[0026] Step S151, the pending target pixel set is divided according to the 256*256 rectangular sub-region division range, and the number of pending targets in each range is screened to complete the first round of false alarm elimination;

[0027] Step S152, the targets in the pending target pixel set after the first round of false alarm elimination are subjected to morphological analysis and screening, the morphological analysis is based on the pending target pixel area and the target contour eccentricity, false targets with larger or smaller contours and approximate circular or too long shapes are eliminated, and the second round of false alarm elimination is completed;

[0028] Step S153, the pending target containing non-background pixels after the second round of false alarm elimination is screened by means of pre-processing mask data, and the third round of false alarm elimination is completed;

[0029] Step S154, the pending target after the third round of false alarm elimination is subjected to background analysis, the background in the contour rectangular range of the pending target is analyzed layer by layer, the clutter characteristics around the pending target are counted, and the real target and cloud debris are distinguished to obtain single-frame target detection data.

[0030] According to one aspect of the application, in step S13, the mean μ z and the variance σ2 Z of the clutter window are calculated according to the following formula:

[0031]

[0032]

[0033] In the formula, N is the total number of pixels of the clutter window, the guard window and the target window, and n is the number of pixels of the clutter window.

[0034] According to one aspect of the application, in step S13, the mean μ z of the clutter window is calculated according to the following formula:

[0035]

[0036] The calculation method of the variance σ2 is as follows:

[0037] The DN value matrix of the original single-band image is constructed, and the DN value dynamic programming summation matrix based on the "Huang Hu Triangle" is constructed. The value of the "Huang Hu Triangle" represents the "area" value from the point to the upper left corner of the single-band image. The rectangular JKLM is the corresponding guard window, and the rectangular IFCG minus the rectangular JKLM is the corresponding clutter window. S represents the DN value statistical characteristics of the window. The statistical characteristic value of the peripheral rectangular IFCG of the clutter window is:

[0038] SIFCG = S ABCD - S ABFH - S AEGD + S AEIH ;

[0039] Statistical characteristic value S of the protection window rectangle JKLM JKLM Statistical characteristic value of the clutter window:

[0040] S IFCG-JKLM = S IFCG - S JKLM .

[0041] According to one aspect of the present application, in the step S2, comprising:

[0042] Step S21, input the single-frame target detection data after false alarm elimination in real time in time sequence, accumulate not less than 5 frames, then continue to input, and do superposition operation on each frame;

[0043] Step S22, do statistical analysis on the superimposed synthesized data set, take the 3rd frame as the main frame, find the potential target within the range of less than 2.5km in distance in the superposition frame for each pending target in the frame, mark the potential target not meeting the target characteristics of the main frame as false alarm, and give the first level of confidence value to the potential target meeting the target characteristics of the main frame;

[0044] Step S23, take the 2nd frame and the 4th frame as the main frame in turn, mark and give the second level of confidence value to the pending target in the superposition frame according to the method in the step S22, eliminate the potential target with lower confidence value and multiple false alarm marks, and generate the initial state scene;

[0045] Step S24, do superposition analysis on the single-frame target detection result added after the 5th frame and the initial state scene, take the single-frame target detection result added as the main frame, find the potential target within the range of less than 2.5km in distance in the initial state scene for each pending target inside, distinguish multiple pending targets within the range, give different confidence values, eliminate the false target with low confidence value, determine the real target position, and calculate the real-time speed and heading information of the target according to the latest target position, the position of the last frame and the time interval;

[0046] Step S25, fuse the single-frame target detection result of the new frame after eliminating false alarm with the initial state scene to generate a new initial state scene, input the single-frame target detection result of the new frame, continue to execute the step four and the step five iteratively, and constantly update the real-time position information and the motion trajectory of the ship target until the input of all original sequence frames ends.

[0047] According to one aspect of the present application, in the step S3, comprising:

[0048] The single-frame target detection result of the newly added frame after false alarm elimination is fused with the initial state scene to generate a new initial state scene, and the single-frame target detection result of a new frame is input, and the iteration of steps S24 and S25 is continuously performed to update the real-time position information and the motion trajectory of the ship target.

[0049] According to an aspect of the present application, an electronic device is provided, comprising one or more processors, one or more memories, and one or more computer programs; wherein the processor is connected with the memory, and the one or more computer programs are stored in the memory, and when the electronic device is running, the processor executes the one or more computer programs stored in the memory, so that the electronic device executes a ship target detection method of single-band image as any one of the above technical solutions.

[0050] According to an aspect of the present application, a computer readable storage medium is provided for storing computer instructions, which are executed by a processor to implement a ship target detection method of single-band image as any one of the above technical solutions.

[0051] According to the concept of the present application, a ship target detection method of single-band image, an electronic device and a computer readable storage medium are provided, by improving the implementation process of the constant false alarm rate detection method, the most time-consuming part of the dynamic programming process processing algorithm with the property of Yang Hui triangle is processed, the preliminary detection result of the ship target is quickly obtained, the preliminary detection result is subjected to four rounds of false alarm elimination by means of the pre-processing mask of the original data, the false alarm rate is reduced as much as possible while the high detection rate is maintained, and the false alarm is further eliminated through the tracking correlation of the sequence frame image, so as to generate the real-time position information and the motion trajectory of the target, which is beneficial to realize the rapid detection of the ship target at sea, and has high time performance and detection performance.

[0052] According to an aspect of the present application, high-orbit satellite single-band sequence images are directly input, preliminary detection and false alarm elimination are performed, the constant false alarm rate detection algorithm is upgraded and improved in the form of a sliding window, real target training model samples are not required to be input, the preliminary detection of single-frame images can be completed, the undetermined target is quickly calibrated to obtain the preliminary detection result and the description information, and a high detection rate is maintained, which does not depend on training samples and can fully utilize single-band images for target detection.

[0053] According to one aspect of the present application, in the single-frame image detection process, after the constant false alarm rate detection algorithm is traversed quickly, the result shows the characteristics of high detection rate and high false alarm rate, the limited features of the pending target are fully utilized, and according to the pre-processing mask segmentation, morphological analysis, target background analysis screening, dynamic tracking correlation and other operations, almost all false alarm targets are eliminated, so as to achieve the effect of detecting real marine ship targets and solve the problem of high false alarm rate in single-band image target detection.

[0054] According to one aspect of the present application, in the detection, it is actually a qualitative process for each pixel of the image. For the input image with a large width, each single-frame image traversal process contains a pixel number of the order of 100 million. The algorithm iterates operation based on the dynamic programming model of the Yang Hui triangle. Without the aid of GPU hardware support, the detection running time of the single-band image containing 10K*10K pixels can be improved to within 20 seconds, solving the problem of long time consumption in the detection process. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 The flowchart schematically shows a ship target detection method of a single-band image according to one embodiment of the present application;

[0056] Figure 2 The flowchart schematically shows a ship target detection method of a single-band image according to another embodiment of the present application;

[0057] Figure 3 The flowchart schematically shows step S1 of the ship target detection method of a single-band image according to one embodiment of the present application;

[0058] Figure 4 The diagram schematically shows the opening of a target window in different regions according to one embodiment of the present application;

[0059] Figure 5 The diagram schematically shows the dynamic planning of the Yang Hui triangle of the image DN value according to one embodiment of the present application;

[0060] Figure 6 The diagram schematically shows the logic operation of a clutter window according to one embodiment of the present application;

[0061] Figure 7 The flowchart schematically shows step S2 of the ship target detection method of a single-band image according to one embodiment of the present application. DETAILED DESCRIPTION

[0062] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0063] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The embodiments cannot be described in detail here, but the embodiments of the present invention are not limited to the following embodiments.

[0064] like Figures 1 to 7 As shown, a ship target detection method for single-band imagery according to the present invention includes the following steps:

[0065] Step S1: Perform fast sliding window detection using single-band image and preprocessed mask as input, and output single-frame target detection results after false alarm rejection;

[0066] Step S2: Based on the motion characteristics of the ship targets and the single-frame target detection results, perform real ship target screening;

[0067] Step S3: Generate the position information and trajectory of the ship target.

[0068] In this embodiment, high-orbit satellite single-band images are directly input, and initial detection and false alarm removal are performed sequentially. The initial detection uses a sliding window approach, and the constant false alarm rate (CFRR) detection algorithm is upgraded and improved. It can complete the initial detection of a single frame image without the input of training model samples of real targets. The target to be detected is quickly calibrated to obtain preliminary detection results and their descriptive information, while maintaining a high detection rate. During the single-frame image detection process, after rapid traversal using the CFRR detection algorithm, the results exhibit the characteristics of high detection rate and high false alarm rate. It makes full use of the limited features of the target to be detected, and based on operations such as preprocessing mask segmentation, morphological analysis, target background analysis and screening, and dynamic tracking and association, almost all false alarm targets are eliminated, so as to achieve the effect of detecting real maritime ship targets. This is conducive to the rapid detection of maritime ship targets and has high time performance and detection performance.

[0069] like Figure 3 and Figure 4 As shown, in one embodiment of the present invention, preferably, step S1 specifically includes:

[0070] Step S11, taking high orbit satellite single band image data as input, obtaining auxiliary information, imaging information and the like of the single band image, and simultaneously carrying out preprocessing operation, performing RPC correction, cloud detection, singular point elimination, sea-land segmentation operation on the image, and fusing the generated various results into a preprocessing mask output, the preprocessing mask is essentially the result of binary segmentation of the original image;

[0071] Step S12, performing double parameter constant false alarm rate detection processing on each pixel of the original single band image, taking the to-be-detected pixel as a target window, taking a rectangular region extending 10 pixels from the center of the target window as a protection window, and taking a ring-shaped region extending 10 pixels to 20 pixels from the center of the target window as a clutter window;

[0072] Step S13, performing DN value characteristic statistics on the original image of the clutter window, and determining a segmentation threshold for target window detection, the opening window range of the target window contains three types relative to the entire image matrix: complete opening window, two-point missing, and three-point missing, and the edge zone of the latter two missing conditions needs to be fully analyzed by using limited pixel DN values, and in the case of a given false alarm rate, the segmentation threshold is determined by the statistical characteristics of the clutter;

[0073] Step S14, performing adaptive threshold pixel classification by using the segmentation threshold to obtain a to-be-determined target pixel set;

[0074] Step S15, obtaining single frame target detection data after four rounds of false alarm elimination on the to-be-determined target pixel set.

[0075] In an embodiment of the present application, preferably, in step S13, the DN value characteristic statistics are performed on the original image of the clutter window, and specifically include:

[0076] The clutter background observation value is set to follow a normal distribution, and the probability density function is:

[0077]

[0078] The detection threshold is set to x=X, and the false alarm probability is p, and the following formula is established:

[0079]

[0080] When x=5, F(5)=0.99999, the detection threshold X is obtained as:

[0081] X=σ·Φ -1 (1-p)+μ

[0082] The above formula is used as the segmentation threshold for target window detection.

[0083] In an embodiment of the present application, preferably, in step S15, specifically includes:

[0084] Step S151, Adaptive partition target upper limit removal: After completing the adaptive threshold pixel classification, a set of target pixels to be determined is obtained. The set of target pixels to be determined is divided into ranges according to the 256×256 rectangular sub-regions. The number of target pixels to be determined in each range is filtered to complete the first round of false alarm removal.

[0085] Step S152, Target Morphological Analysis and Removal: Morphological analysis and screening are performed on the targets in the target pixel set after the first round of false alarm removal. The morphological analysis is based on the target pixel area and target contour eccentricity. False targets with large or small contours, or those that are approximately circular or too long and narrow are removed to complete the second round of false alarm removal.

[0086] Step S153, Cloud Mask Coverage Removal: Using preprocessed mask data, the target to be determined containing non-background pixels after the second round of false alarm removal is screened to complete the third round of false alarm removal.

[0087] Step S154, Adaptive Background Analysis and Removal: Perform background analysis on the target to be identified after the third round of false alarm removal. Analyze the background within the rectangle of the target's outline layer by layer, statistically amplify the clutter characteristics around the target, distinguish between real targets and fragmented clouds, and obtain single-frame target detection data.

[0088] In one embodiment of the present invention, preferably, when statistically analyzing the DN value characteristics of the clutter window containing each target window, directly calculating the mean and variance is time-consuming and involves repetitive calculations. Therefore, in processing this part, the calculation process of the variance and mean is accelerated, based on the calculation formula of the mean and variance method:

[0089]

[0090]

[0091] In step S13, the mean value μ of the clutter window is calculated. z and variance The calculation formula is:

[0092]

[0093]

[0094] In the formula, N is the total number of pixels in the clutter window, guard window and target window, and n is the number of pixels in the clutter window.

[0095] In one embodiment of the present invention, preferably, in step S13, the mean value μ of the clutter window is... z The calculation formula is:

[0096]

[0097] variance The calculation method is as follows:

[0098] The Pascal's Triangle has a property that "each number is equal to the sum of the two numbers above it". This property was transplanted into the original image matrix and adaptively improved. The entire traversal process was decomposed into smaller summation subproblems. Each subproblem was solved one by one and the results were stored in an array. Therefore, each subproblem was calculated only once, reducing the nesting level and thus greatly improving the running speed.

[0099] The principle and process are as follows Figure 5 , Figure 6 As shown, Figure 5 (a) is an example of the original image DN value matrix, and its corresponding region is Figure 6 Rectangle ABCD in the middle; Figure 5 (b) Dynamic programming is used to find the summation matrix for the DN value of the improved Pascal's Triangle property. Solid arrows represent addition operations, and dashed arrows represent subtraction operations. Each value is obtained according to the operational rules of the arrows in the figure. The value represents the "area" of the rectangle from that point to the upper left corner of the image. Figure 5 In (b), "112" represents Figure 6 The "area" attribute value of rectangle ABCD in the image is obtained. After traversing the entire image matrix in this way, the "area" attribute value defined by any pixel position can be obtained. Then, after performing coverage surface logic operations, the DN value characteristics of the clutter window where each target window is located are obtained. The coverage surface logic operation process is as follows: Figure 6 As shown, the black rectangle represents the target window, rectangle JKLM represents the corresponding guard window, and rectangle IFCG minus rectangle JKLM represents the corresponding clutter window. Let S represent the statistical characteristics of the window's DN values ​​(sum of ND values, sum of squared ND values). Then, the statistical characteristic value of the outer rectangle IFCG of the clutter window is:

[0100] S IFCG =S ABCD -S ABFH -S AEGD +S AEIH

[0101] The corresponding pixel calculation relationship is as follows:

[0102] S IFCG =S4-S2-S3+S1

[0103] Following the above logical process, the statistical characteristic value S of the guard window rectangle JKLM is also obtained. JKLM Finally, the statistical characteristics of the clutter window (rectangular IFCG - rectangular JKLM) can be calculated:

[0104] S IFCG-JKLM =S IFCG -S JKLM ;

[0105] In this way, the calculation process of large matrix mean and variance containing hundreds of millions of pixels can be quickly completed, thereby improving the time efficiency of the entire detection process.

[0106] As Figure 7 shown in the embodiment of the present application, preferably, under the premise of obtaining single-frame target detection data, the pending ship targets in the single frame still contain part of false alarms, in step S2, comprising:

[0107] Step S21, input the single-frame target detection data after false alarm elimination in time sequence in real time, accumulate not less than 5 frames, then continue to input, and do superposition operation for each frame;

[0108] Step S22, do statistical analysis on the superimposed synthetic data set, take the 3rd frame as the main frame, find the potential target within the range of less than 2.5km in distance in the superimposed frame for each pending target in the frame, mark the potential target not meeting the target characteristics of the main frame as false alarm, and give the first level of confidence value to the potential target meeting the target characteristics of the main frame;

[0109] Step S23, take the 2nd frame and the 4th frame as the main frame in turn, mark and give the second level of confidence value to the pending target in the superimposed frame according to the method in step S22, eliminate the potential target with lower confidence and multiple false alarm marks, and generate the initial state scene;

[0110] Step S24, do superposition analysis on the single-frame target detection result added after the 5th frame and the initial state scene, take the single-frame target detection result added as the main frame, find the potential target within the range of less than 2.5km in distance in the initial state scene for each pending target inside, distinguish multiple pending targets in the range, give different confidence values, eliminate the false targets with low confidence, determine the real target position, and calculate the real-time speed and heading information of the target according to the latest target position, the position of the last frame and the time interval;

[0111] Step S25, fuse the single-frame target detection result of the new frame after eliminating false alarms and the initial state scene to generate a new initial state scene, input the single-frame target detection result of the new frame, continue to execute step four and step five iteratively, and constantly update the real-time position information and motion trajectory of the ship target until all the original sequence frames are inputted.

[0112] In the embodiment of the present application, preferably, in step S3, comprising:

[0113] Fuse the single-frame target detection result of the new frame after false alarm elimination with the initial state scene to generate a new initial state scene, input the single-frame target detection result of a new frame, and continue to perform iteration according to steps S24 and S25 to update the real-time position information and motion trajectory of the ship target.

[0114] According to an aspect of the present application, an electronic device is provided, comprising one or more processors, one or more memories, and one or more computer programs; wherein the processor is connected with the memory, and the one or more computer programs are stored in the memory, and when the electronic device is running, the processor executes the one or more computer programs stored in the memory to enable the electronic device to perform the ship target detection method of the single-band image according to any one of the above technical solutions.

[0115] According to an aspect of the present application, a computer readable storage medium is provided for storing computer instructions, and when the computer instructions are executed by a processor, a ship target detection method of a single-band image according to any one of the above technical solutions is implemented.

[0116] To sum up, the present application provides a ship target detection method of a single-band image, an electronic device and a computer readable storage medium, which directly inputs high-orbit satellite single-band images, sequentially performs preliminary detection and false alarm elimination, upgrades the constant false alarm rate detection algorithm in the form of a sliding window, does not need to input real target training model samples, and can complete the preliminary detection of a single frame of image, quickly calibrates the pending target to obtain the preliminary detection result and its description information, and maintains a high detection rate. In the single frame image detection process, after the constant false alarm rate detection algorithm is quickly traversed, the result shows the characteristics of high detection rate and high false alarm rate, fully utilizes the limited features of the pending target, and almost eliminates all false alarm targets through pre-processing mask segmentation, morphological analysis, target background analysis screening, dynamic tracking correlation and other operations, so as to achieve the effect of detecting real sea ship targets, which is beneficial to realize the rapid detection of sea ship targets and has high time performance and detection performance.

[0117] In addition, it should be noted that the present application can be provided as a method, device or computer program product. Therefore, the embodiments of the present application can adopt a completely hardware embodiment, a completely software embodiment or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present application can adopt the form of a computer program product implemented on one or more computer usable storage media containing computer usable program codes.

[0118] The embodiments of the present application are described with reference to the flowchart illustrations and / or block diagrams of the methods, terminal devices (systems) and computer program products according to embodiments of the present application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing terminal devices to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal devices, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0119] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal devices to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means that implement the function specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0120] It should also be noted that, in the present document, the terms "comprising", "containing" or any other variant thereof are intended to cover non-exclusive inclusions, such that a process, method, article or terminal device that includes a list of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or terminal device. Without more limitations, an element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the process, method, article or terminal device that includes the said element.

[0121] Finally, it should be noted that the above description is of preferred embodiments of the application, and that although preferred embodiments of the application have been described, numerous changes and modifications can be made to the preferred embodiments without departing from the principles of the application, and that such changes and modifications are contemplated as falling within the scope of the application. Accordingly, the appended claims are intended to embrace all such changes and modifications.

Claims

1. A ship target detection method of single-band image, comprising the following steps: Step S1, performing fast sliding window detection with single-band image and preprocessed mask as input, and outputting single-frame target detection result after false alarm elimination; Step S2, performing real ship target screening according to the motion characteristics of ship target and the single-frame target detection result; Step S3, generating position information and motion trajectory of ship target; In the step S1, specifically comprising: Step S11, taking high-orbit satellite single-band image data as input, obtaining auxiliary information and imaging information of single-band image; Step S12, performing double-parameter constant false alarm rate detection processing on each pixel of original single-band image, taking the to-be-detected pixel as target window, taking the rectangle region extended by 10 pixels from the center of target window as protection window, and taking the ring region extended by 10 pixels to 20 pixels from the center of target window as clutter window; Step S13, performing DN value characteristic statistics on original image of clutter window, and determining the segmentation threshold of target window detection; Step S14, completing adaptive threshold pixel classification by using the segmentation threshold, and obtaining the set of to-be-determined target pixels; Step S15, obtaining single-frame target detection data after four rounds of false alarm elimination on the set of to-be-determined target pixels; In the step S13, the DN value characteristic statistics on the original image of the clutter window specifically comprises: The clutter background observation value is subject to normal distribution, and the probability density function is: Setting the detection threshold to , the false alarm probability to , the following holds: When , , the detection threshold X is obtained as: The above formula is used as the segmentation threshold of target window detection; In step S13, the mean value of the clutter window The calculation formula is: Variance The calculation method is: The DN value matrix example rectangle ABCD of original single-band image and the DN value dynamic programming summation matrix based on "Huang Huili triangle" are constructed, the value of "Huang Huili triangle" represents the "area" value from the point to the upper left corner of single-band image, rectangle JKLM is the corresponding protection window, and the statistical characteristic value of the DN value of the outer rectangle IFCG of the clutter window is: ； statistical property value of the protection window rectangle JKLM statistical property value of the clutter window 。 2. The method of claim 1, wherein, In the step S15, specifically comprising: Step S151, dividing the set of to-be-determined target pixels into 256*256 rectangular sub-regions, screening the number of to-be-determined targets in each range, and completing the first round of false alarm elimination; Step S152, performing morphological analysis and screening on the targets in the set of to-be-determined target pixels after the first round of false alarm elimination, and completing the second round of false alarm elimination according to the to-be-determined target pixel area and target profile eccentricity; Step S153, screening the to-be-determined targets containing non-background pixels after the second round of false alarm elimination by means of preprocessed mask data, and completing the third round of false alarm elimination; Step S154, performing background analysis on the to-be-determined targets after the third round of false alarm elimination, statistically analyzing the background in the to-be-determined target profile rectangular range layer by layer, amplifying the clutter characteristics around the to-be-determined target, and distinguishing the real target from the broken cloud, to obtain single-frame target detection data.

3. The method of claim 1, wherein, In step S13, the mean value of the clutter window and the formula for calculating the variance are In the formula, N is the total number of pixels of the clutter window, the protection window and the target window, and n is the number of pixels of the clutter window.

4. The method of claim 1, wherein, In the step S2, comprising: Step S21, input the single-frame target detection data after false alarm elimination in real time in time sequence, accumulate not less than 5 frames, then continue to input, and do superposition operation for each frame; Step S22, do statistical analysis on the superimposed synthetic data set, take the 3rd frame as the main frame, find the potential target within the range of less than 2.5km in the superimposed frame for each pending target in the main frame, mark the potential target which does not meet the target characteristics of the main frame as false alarm, and give the first level of credibility to the potential target which meets the target characteristics of the main frame; Step S23, take the 2nd frame and the 4th frame as the main frame in turn, mark and give the second level of credibility to the pending target in the superimposed frame according to the method in step S22, eliminate the potential target which has lower credibility and is marked as false alarm for many times, and generate the initial state scene; Step S24, do superposition analysis on the single-frame target detection result added after the 5th frame and the initial state scene, take the single-frame target detection result added as the main frame, find the potential target within the range of less than 2.5km in the initial state scene for each pending target in the main frame, distinguish the multiple pending targets within the range, give different credibility, eliminate the false target with low credibility, determine the real target position, and calculate the real-time speed and heading information of the target according to the latest target position, the position of the last frame and the time interval; Step S25, fuse the single-frame target detection result of the new frame after eliminating false alarm with the initial state scene to generate a new initial state scene, input the single-frame target detection result of the new frame, continue to execute steps four and five iteratively, and update the real-time position information and motion trajectory of the ship target until all original sequence frames are inputted.

5. The method of claim 4, wherein, In the step S3, it includes: Fuse the single-frame target detection result of the new frame after eliminating false alarm with the initial state scene to generate a new initial state scene, input the single-frame target detection result of the new frame, continue to execute steps S24 and S25 iteratively, and update the real-time position information and motion trajectory of the ship target.

6. An electronic device, comprising: It includes: One or more processors, one or more memories, and one or more computer programs; wherein the processor is connected with the memory, the one or more computer programs are stored in the memory, when the electronic device is running, the processor executes the one or more computer programs stored in the memory, so that the electronic device executes the ship target detection method of single-band image as claimed in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that, A computer program for storing computer instructions, when executed by a processor, implements the ship target detection method of single-band image as claimed in any one of claims 1 to 5.

Images