Ship target key part identification method and system

Abstract

The invention discloses a ship target key part identification method and system, and belongs to the field of key part identification, and the method comprises the steps: obtaining a radiometer image of a ship target; segmenting a ship target from the radiometer image, sampling a contour curve or a boundary of the ship target in a segmentation result to obtain a target contour sampling point set, and extracting shape features of each contour sampling point in the target contour sampling point set; calculating the similarity between the shape feature of each contour sampling point in the target contour sampling point set and the shape feature of each contour sampling point of the template ship to obtain a similarity matrix; the template ship is a known ship of each key part; and according to the similarity matrix, screening out part of contour sampling points having the highest similarity with each key part of the template ship from the target contour sampling point set so as to position the area where the corresponding key part in the ship target is located. According to the method, the key part of the ship target can be effectively identified according to the radiometer image of the ship target, so that the radiometer is fully utilized to detect good performance.

Description

Technical Field

[0001] The present invention belongs to the field of key part identification, and more specifically, relates to a method and system for identifying key parts of a ship target. Background Art

[0002] With the rapid development of high-tech, the form of maritime warfare is constantly evolving, and intelligence, miniaturization, unmanned operation and precision have become the trend of maritime warfare. However, relying solely on target point attacks may not cause devastating blows to target ships. Using image detection technology to obtain ship images and using image processing technology to process ship detection images to identify key parts of the ship such as the bow, island, and stern as attack targets is of great significance in maritime combat scenarios.

[0003] In recent years, most methods for identifying key parts of ships are based on infrared images or satellite remote sensing images. For example, in the patent document with application publication number CN117095159A, an intelligent identification method for key parts of ships in infrared images is disclosed. The patent proposes an infrared ship detection network model, which automatically detects and obtains the characteristic information of the ship (including the bow, stern, length and the central axis of the ship) by acquiring infrared detection images. The patent constructs an infrared ship detection network model of improved YOLOv7 to detect the ship target and its key parts information. For another example, in the patent document with authorization announcement number CN107563447B, a hierarchical identification method for target parts based on remote sensing images is disclosed. The patent extracts features from the pixel points of the target area of ​​the ship, then constructs a support tensor machine or a support vector machine for training, and uses the trained model to classify the pixel points of the ship area, thereby realizing the identification of key parts.

[0004] Different from infrared images or satellite remote sensing images, microwave radiometers use electromagnetic signals emitted or reflected by the target itself to detect the target scene. Since radiometer detection has the advantages of good concealment, passive detection, all-day and all-weather, it has gradually become an important equipment for sea surface ship detection. However, compared with infrared images or remote sensing images, the sea surface scene image quality detected by microwave radiometers is poor, the gradient change of target contours is not obvious, and the clear contour of the internal area of ​​the ship cannot be obtained, and the key parts cannot be directly distinguished. The above-mentioned ship target key part recognition methods all need to complete the division of internal key parts first, which makes the ship target radiometer image unable to adapt to the existing ship target key part recognition methods, limiting the application of microwave radiometer images in ship target key part recognition. In addition, it is difficult to construct a data set for microwave radiometer images. Therefore, it is impossible to use neural networks or vector machines to train a model for the recognition of ship key parts.

[0005] In general, how to identify the key parts of ship targets based on the radiometer images of ship targets is still a difficult problem to be solved. Summary of the invention

[0006] In view of the defects of the prior art and the need for improvement, the present invention provides a method and system for identifying key parts of ship targets. The purpose is to propose a method that can effectively identify key parts of ship targets based on the radiometer image of the ship target, so as to make full use of the advantages of radiometer detection such as good concealment, passive detection, all-day and all-weather detection, etc.

[0007] To achieve the above object, according to one aspect of the present invention, a method for identifying key parts of a ship target is provided, comprising:

[0008] S1: Acquire the radiometer image of the ship target;

[0009] S2: Segment the ship target from the radiometer image, and sample the contour curve or boundary of the ship target in the segmentation result to obtain a target contour sampling point set;

[0010] S3: extracting shape features of each contour sampling point in the target contour sampling point set;

[0011] S4: Calculate the similarity between the shape features of each contour sampling point in the target contour sampling point set and the shape features of each contour sampling point of the template ship to obtain a similarity matrix; the template ship is a ship with known key parts;

[0012] S5: For each key part of the template ship, the contour sampling points of the key part are used as the matching template, and the partial contour sampling points with the highest similarity to each matching template are selected from the target contour sampling point set according to the similarity matrix to locate the area where the corresponding key part of the ship target is located.

[0013] In some optional embodiments, the shape feature of the contour sampling point is the x-coordinate or y-coordinate of the contour sampling point.

[0014] Furthermore, in step S4, a dynamic time warping DTW algorithm or a FastDTW algorithm is used to calculate the similarity between the shape features of each contour sampling point in the target contour sampling point set and the shape features of each contour sampling point of the template ship.

[0015] Furthermore, in step S2, full sampling is used to sample the contour curve or boundary of the ship target in the segmentation result.

[0016] In some optional embodiments, the shape features of the contour sampling points are shape context histogram features or directional gradient histogram features.

[0017] Furthermore, in step S4, the cosine similarity is used to calculate the similarity between the shape features of each contour sampling point in the target contour sampling point set and the shape features of each contour sampling point of the template ship.

[0018] Furthermore, in step S2, Jitendra sampling or uniform sampling is used to sample the contour curve or boundary of the ship target in the segmentation result.

[0019] Furthermore, before extracting the shape features of each contour sampling point in the target contour sampling point set, the method further includes: correcting the orientation angle of the ship target so that the orientation angle of the ship target is the same as the orientation angle of the template ship.

[0020] Furthermore, the heading angle of the ship target is corrected, including:

[0021] Perform ellipse fitting on the contour sampling points of the ship target;

[0022] Estimate the direction angle of the ship target according to the ellipse fitting result;

[0023] Rotates the ship target's heading to match the template ship's heading.

[0024] According to another aspect of the present invention, a ship target key part identification system is provided, characterized in that it includes:

[0025] A detection module, used to obtain a radiometer image of a ship target;

[0026] The segmentation module is used to segment the ship target from the radiometer image, and to sample the contour curve or boundary of the ship target in the segmentation result to obtain a target contour sampling point set;

[0027] A feature extraction module, used to extract shape features of each contour sampling point in the target contour sampling point set;

[0028] A similarity calculation module is used to calculate the similarity between the shape features of each contour sampling point in the target contour sampling point set and the shape features of each contour sampling point of the template ship to obtain a similarity matrix; the template ship is a ship with known key parts;

[0029] And the matching module, for each key part of the template ship, takes the contour sampling points of the key part as the matching template, and selects the part of the contour sampling points with the highest similarity to each matching template from the target contour sampling point set according to the similarity matrix to locate the area where the corresponding key part of the ship target is located.

[0030] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects:

[0031] (1) The present invention first obtains the radiometer image of the ship target as the basis for identifying the key parts of the ship target, and realizes the identification of the key parts of the ship target based on the idea of ​​template matching. Specifically, the ship target and the sea background are separated by image segmentation, and then the contour sampling points are obtained and the shape features of the contour sampling points are extracted. The similarity matrix is ​​calculated by calculating the feature similarity between the contour sampling points of the ship target and the contour sampling points of the template ship, and the key parts of the ship target are located based on the similarity matrix. When identifying the key parts of the ship target, the present invention does not need to distinguish different key parts in advance, nor does it need to construct a data set. For radiometer images that cannot clearly detect the internal details of the target, the key parts of the ship can also be accurately identified. Compared with the existing ship key part identification method based on infrared images or satellite remote sensing images, the present invention can effectively utilize the advantages of radiometer detection such as good concealment, passive detection, all-day and all-weather, etc., to improve the effectiveness of attacking marine ship targets.

[0032] (2) There is no internal structural feature inside the radiometer image. The present invention extracts the shape features of the contour sampling points as the basis for matching, which is independent of the internal structural features of the image and can also avoid interference from internal noise. In some optional embodiments of the present invention, the relative coordinate position (i.e., x coordinate or y coordinate) of the contour sampling points is used as the shape feature of each contour sampling point, which can effectively reduce the computational complexity of feature extraction. On this basis, the dynamic time warping DTW algorithm or the FastDTW algorithm is used to calculate the similarity between the shape features of the contour sampling points. On the one hand, the global feature information can be comprehensively considered, and on the other hand, the computational efficiency of the overall matching can be effectively improved; in some other optional embodiments of the present invention, the shape context histogram feature or the directional gradient histogram feature is used as the shape feature of each contour sampling point. In the process of feature extraction, the global information is considered, ensuring that the features of each contour sampling point at different scales are consistent, avoiding the influence of the multi-scale problem of the ship target on the feature extraction during the radiometer image detection process, and further improving the recognition accuracy of the key parts of the ship target.

[0033] (3) When the present invention uses the relative coordinate position (i.e., x-coordinate or y-coordinate) of the contour sampling point as the shape feature of each contour sampling point, full sampling is adopted to obtain the contour sampling points of the ship target, thereby retaining all contour points. Since the computational complexity of feature extraction and similarity calculation is low, the present invention can retain relatively complete contour information while ensuring a high recognition efficiency, thereby further improving the accuracy of key part recognition.

[0034] (4) When the present invention uses shape context histogram features or directional gradient histogram features as shape features of contour sampling points, Jitendra sampling or uniform sampling is used to obtain contour sampling points of ship targets, thereby achieving relatively uniform sampling and reducing the amount of computation for feature extraction and similarity calculation while retaining key information.

[0035] (5) In the preferred embodiment of the present invention, before extracting the shape features of the contour sampling points, the orientation angle of the ship target will be corrected to make it consistent with the orientation angle of the template ship, thereby ensuring that the features of the same contour sampling point at different orientation angles are consistent, avoiding the influence of the multi-angle problem of the ship target on the feature extraction during the radiometer image detection process, and further improving the recognition accuracy of the key parts of the ship target. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flow chart of a method for identifying key parts of a ship target provided in Example 1 of the present invention;

[0037] Figure 2 A schematic diagram of a radiometer image provided in Example 1 of the present invention;

[0038] Figure 3 A schematic diagram of sampling points of the contour of a ship target provided in Example 1 of the present invention;

[0039] Figure 4 A schematic diagram of the ellipse fitting results of the sampling points of the ship target contour provided in Example 1 of the present invention;

[0040] Figure 5 A schematic diagram of feature extraction of sampling points of ship target contour provided in Example 1 of the present invention;

[0041] Figure 6 The template ship image provided in Example 1 of the present invention;

[0042] Figure 7 A schematic diagram of a similarity matrix of template ship contour sampling points and target ship contour sampling points provided in Example 1 of the present invention;

[0043] Figure 8 A schematic diagram of the matching results of the key parts of the template ship and the key parts of the target ship provided in Example 1 of the present invention;

[0044] Fig. 9 A schematic diagram of sampling points of a template ship outline provided in Example 2 of the present invention;

[0045] Fig.10A schematic diagram of sampling points of a ship target contour provided in Example 2 of the present invention; wherein (a) is a schematic diagram of sampling points of an original ship target contour, and (b) is a schematic diagram of sampling points of a ship target contour after orientation angle correction;

[0046] Fig.11 A schematic diagram of shape features provided in Example 2 of the present invention; wherein (a) is the x-coordinate sequence of the sampling points of the template ship contour, (b) is the y-coordinate sequence of the sampling points of the template ship contour, (c) is the x-coordinate sequence of the sampling points of the ship target contour, and (d) is the y-coordinate sequence of the sampling points of the ship target contour;

[0047] Fig.12 A schematic diagram of a similarity matrix provided in Example 2 of the present invention;

[0048] Fig.13 Schematic diagram of the key parts identification results of a ship target provided in Example 2 of the present invention; wherein (a) is a schematic diagram of the bow identification results, (b) is a schematic diagram of the stern identification results, and (c) is a schematic diagram of the island identification results. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0050] In the present invention, the terms "first", "second", etc. (if any) in the present invention and the accompanying drawings are used to distinguish similar objects but not necessarily to describe a specific order or sequence.

[0051] In order to utilize the advantages of radiometer detection such as good concealment, passive detection, all-day and all-weather, etc. to improve the effectiveness of attacking maritime ship targets in maritime combat, the present invention provides a method and system for identifying key parts of ship targets. The overall idea is to realize the identification of key parts of ship targets through template matching, in view of the fact that radiometer images cannot clearly detect the internal details of the target, thereby avoiding dependence on the internal details of the target and the data set.

[0052] The following are examples.

[0053] Embodiment 1:

[0054] A method for identifying key parts of a ship target, such as Figure 1 As shown, including:

[0055] S1: Acquire the radiometer image of the ship target;

[0056] S2: Segment the ship target from the radiometer image, and sample the contour curve or boundary of the ship target in the segmentation result to obtain a target contour sampling point set;

[0057] S3: extracting shape features of each contour sampling point in the target contour sampling point set;

[0058] S4: Calculate the similarity between the shape features of each contour sampling point in the target contour sampling point set and the shape features of each contour sampling point of the template ship to obtain a similarity matrix; the template ship is a ship with known key parts;

[0059] S5: For each key part of the template ship, the contour sampling points of the key part are used as the matching template, and the partial contour sampling points with the highest similarity to each matching template are selected from the target contour sampling point set according to the similarity matrix to locate the area where the corresponding key part of the ship target is located.

[0060] The specific implementation of each step is further explained below.

[0061] In practical applications, the radiometer system for acquiring the radiometer image of the ship target may adopt a real aperture radiometer, a synthetic aperture radiometer, etc.; optionally, in step S1 of this embodiment, a synthetic aperture radiometer is adopted. Figure 2 Shown is an example of a radiometer image.

[0062] In practical applications, the algorithm for segmenting the ship target from the radiometer image can adopt a threshold segmentation algorithm, an edge detection algorithm, a segmentation algorithm based on regions and boundaries, and the like; after the ship target is segmented, the contour sampling points are obtained by sampling, so that the amount of calculation and the calculation time in the process of identifying the key parts of the ship target can be controlled. Specifically, full sampling, Jitendra sampling, uniform sampling, random sampling, equal interval sampling, and the like can be used; optionally, in step S2 of this embodiment, an edge detection algorithm is specifically used to achieve the segmentation of the ship target; in addition, after obtaining the contour sampling points, this embodiment needs to calculate the shape context histogram features of the contour sampling points. The calculation complexity of the shape feature extraction process is relatively high. Considering that the internal detail information of the target in the radiometer image is unknown, in order to ensure the accuracy of the key part identification, the uniformity of the sampling needs to be ensured. As a preferred implementation mode, the sampling method selected in this embodiment is specifically Jitendra sampling. Through this sampling method, a number of representative pixel points can be selected from the contour curve or boundary of the separated ship target. The result of the ship contour sampling points is as follows: Figure 3 As shown, these points can better describe the geometric characteristics of the contour, such as shape and structure, and reduce the amount of data while retaining as much original contour information as possible.

[0063] There is a multi-angle problem in the process of radiometer image detection, that is, when the same contour sampling point is at different orientation angles of the target, the extracted features are inconsistent. Traditional template matching cannot effectively solve the problem of multi-angle matching. In order to avoid the influence of the multi-angle problem on the matching result, as a preferred implementation, in step S3 of this embodiment, before extracting the shape features of each contour sampling point in the target contour sampling point set, it also includes: correcting the orientation angle of the ship target so that the orientation angle of the ship target is the same as the orientation angle of the template ship; specifically, this embodiment combines the shape characteristics of the ship target to perform ellipse fitting on its contour sampling points, such as Figure 4 As shown, the principle of contour point ellipse fitting is as follows:

[0064] The equation of the known ellipse is:

[0065] ax 2 +bxy+cy 2 +dx+ey=1

[0066] Among them, a, b, c, d, e are fitting coefficients; let A = [a, b, c, d, e] T , X=[x 2 ,xy,y 2 ,x,y] T , so the equation can be expressed as AX = 1. Then the optimization problem of fitting the ellipse can be expressed as:

[0067] min||DA|| 2

[0068] sA T CA=1

[0069] Where D represents the data sample set N×6, 6 represents the dimension, and N represents the number of contour sampling points. A represents the parameters of the ellipse equation, and the constant matrix C is:

[0070]

[0071] After solving the ellipse equation of the ship's contour points according to the above optimization equation, it is necessary to solve the ship's heading angle. The estimation equation of the ship's heading angle is as follows:

[0072]

[0073] The N contour sampling points of the target ship are rotated in reverse order to make the angles of the target ship and the template ship consistent. The equation for the contour point rotation transformation is as follows:

[0074]

[0075] Among them, (x, y) are the coordinates of the target ship's contour sampling points; θ is the ship's heading angle; (x1, y1) are the coordinates of the ship's contour sampling points after angle correction.

[0076] During the radiometer detection process, the size of the ship target in the radiometer image changes due to the change in the observation distance. Therefore, in order to solve the multi-scale problem in the ship matching process, it is necessary to construct a feature vector for each contour sampling point. As a preferred implementation, in step S3 of this embodiment, the shape feature of the extracted contour sampling point is specifically a shape context histogram feature, such as Figure 5 As shown in the figure, when extracting the shape context histogram features of each contour sampling point, for each contour point, a polar logarithmic coordinate system centered on it is constructed (including K1 angle regions and K2 distance regions), and then the contour points around it are mapped into each region, and then the number of contour points falling in each region is counted, and finally normalized (i.e. divided by the number of contour points falling in all regions), thus generating a shape context histogram matrix. The specific construction method of the shape context histogram feature includes:

[0077] Calculate the distance matrix between point pairs and divide it by the mean distance to make the shape context descriptor insensitive to scaling;

[0078] The distance matrix is ​​mapped to the logarithmic domain and compared with the threshold, thereby dividing five distance segments in the logarithmic domain. The distance matrix and angle matrix between the contour sampling points are solved as follows:

[0079]

[0080] Among them, p and q are two contour sampling points of the ship target; (x p ,y p ) represents the coordinates of the contour sampling point p, (x q ,y q ) represents the coordinates of the contour sampling point q, disp(p,q) represents the distance between the contour sampling points p and q, Represents the angle between contour sampling points p and q;

[0081] After solving the angle matrix and the distance matrix, region mapping is performed to count the number of points in the K1×K2 regions. Each contour sampling point can finally be represented by a K-dimensional feature vector, where the feature vector dimension K=K1×K2.

[0082] It should be noted that the shape context histogram feature is only a preferred implementation of the present invention and should not be understood as the only limitation to the present invention. In some other embodiments of the present invention, other shape features such as directional gradient histogram features may also be used.

[0083] In practical applications, any ship with known internal key parts can be selected as a template ship. Figure 6 The template ship image selected in this embodiment is shown. According to the same feature extraction method, the contour sampling points of the template ship and the K-dimensional feature vectors of each contour sampling point can be extracted; the number of contour sampling points in the template ship and the ship target is recorded as M and N respectively, and the similarity of the feature vectors between the two groups of contour points is calculated to obtain an M×N similarity matrix, as shown in FIG. Figure 7 As shown, P1-P5 represent the feature vectors of each contour sampling point in the template ship, P'1-P'7 represent the feature vectors of each contour sampling point in the ship target, and each element value in the matrix represents the similarity between two feature vectors; the similarity between feature vectors can be calculated by cosine similarity, Euclidean distance, etc. Optionally, in this embodiment, cosine similarity is used for calculation, and the calculation formula is as follows:

[0084]

[0085] Among them, P and Q are the feature vectors of two contour sampling points respectively; p i is the i-th eigenvalue of the sampling point eigenvector P; q i is the i-th eigenvalue of the sampling point feature vector Q; cosγ represents the cosine similarity between the feature vectors P and Q.

[0086] Based on the calculated similarity matrix, contour sampling points of key parts such as the bow, island, and stern of the template ship are selected respectively, such as Figure 6 As shown, according to the similarity matrix, the contour sampling points with the greatest similarity to the contour sampling points of these key parts are found from the target contour sampling points of the ship target, so as to obtain the key parts area of ​​the ship target in the radiometer detection scene, such as Figure 8 It is easy to understand that the similarity between shape features is related to the similarity measurement relationship. When cosine similarity is used, the greater the cosine similarity, the greater the similarity between shape features.

[0087] In general, this embodiment provides a method for identifying key parts of ship targets based on radiometer images and template matching. The method first detects the radiometer image of the ship target, and then separates the ship target from the sea background; in view of the multi-angle problem of the ship target during the detection process, the ship angle correction is performed by the ellipse fitting method; in view of the multi-scale problem of the ship target during the detection process, the shape context histogram features of the contour sampling points are constructed to eliminate the scale differences under different observation distances; finally, the matching and positioning of the key parts area is achieved by calculating the similarity measurement matrix of the target ship contour points and the template ship contour points. Based on the key part identification method, this embodiment can be combined with a detection system such as a radiometer that cannot clearly detect the internal details of the target, and does not need to build a data set or pre-train model parameters, and can accurately distinguish the key parts of the ship such as the bow, island, and stern, making it possible to use the advantages of good concealment, passive detection, all-day and all-weather detection of the radiometer to improve the effectiveness of attacking marine ship targets.

[0088] Embodiment 2:

[0089] A method for identifying key parts of a ship target. This embodiment is similar to the above-mentioned embodiment 1, and includes the following steps S1 to S5:

[0090] S1: Acquire the radiometer image of the ship target;

[0091] S2: Segment the ship target from the radiometer image, and sample the contour curve or boundary of the ship target in the segmentation result to obtain a target contour sampling point set;

[0092] S3: extracting shape features of each contour sampling point in the target contour sampling point set;

[0093] S4: Calculate the similarity between the shape features of each contour sampling point in the target contour sampling point set and the shape features of each contour sampling point of the template ship to obtain a similarity matrix; the template ship is a ship with known key parts;

[0094] S5: For each key part of the template ship, the contour sampling points of the key part are used as the matching template, and the partial contour sampling points with the highest similarity to each matching template are selected from the target contour sampling point set according to the similarity matrix to locate the area where the corresponding key part of the ship target is located.

[0095] The difference between this embodiment and the above embodiment is that in step S3 of this embodiment, the shape feature of the extracted contour sampling point is the x-coordinate of the contour sampling point. Correspondingly, in step S4, the dynamic time warping DTW algorithm or the FastDTW algorithm is used to calculate the similarity between the shape features of the contour sampling points. Compared with the shape context histogram features extracted in the above embodiment 1, in this embodiment, the calculation complexity of feature extraction of each contour sampling point is significantly reduced. When using the dynamic time warping DTW algorithm or the FastDTW algorithm to calculate the similarity between shape features, global feature information can be considered, and the overall complexity of feature extraction and similarity calculation is significantly reduced.

[0096] Since the overall complexity of feature extraction and similarity calculation is relatively low, in order to further improve the accuracy of key part recognition of the ship, in step S2 of this embodiment, full sampling is used to sample the contour curve or boundary of the ship target in the segmentation result. In this way, relatively complete contour information can be retained while ensuring a high recognition efficiency, further improving the accuracy of key part recognition.

[0097] Similar to the above-mentioned embodiment 1, in order to solve the multi-angle problem in radiometer image detection, this embodiment will also perform ellipse fitting and orientation angle correction based on the shape characteristics of the ship target before contour sampling, which is essentially consistent with the orientation angle of the template ship.

[0098] Fig. 9 FIG. 1 is a schematic diagram of the outline of the template ship selected in this embodiment; Fig.10 (a) in FIG. 1 is a schematic diagram of the outline of the ship target after extraction and full sampling in this embodiment. Fig.10 (b) in the figure shows a schematic diagram of the outline of the ship target after the heading angle correction.

[0099] Record the x- and y-coordinates of each contour sampling point in the template ship and the ship target in the same direction (clockwise or counterclockwise), and obtain Fig.11 The coordinate sequence shown, Fig.11 (a) and (b) represent the x-coordinate sequence and y-coordinate sequence of the template ship, respectively. Fig.11 (c) and (d) in the figure represent the x-coordinate sequence and y-coordinate sequence of the ship target respectively.

[0100] Select Fig.11 (a) and (c) in the figure are used as the shape feature sequences of the template ship contour sampling points and the ship target contour sampling points. The FastDTW algorithm is used to calculate the similarity between the shape features, and the state transfer matrix is ​​obtained as follows: Fig.12 As shown in the figure, by using the state transfer matrix as the similarity matrix, the key parts of the ship target can be identified and located. Fig.13 (a), (b) and (c) are schematic diagrams of the recognition results of the bow, stern and island of the ship in this embodiment.

[0101] It is easy to understand that in some other embodiments of the present invention, the y coordinate can also be used as the shape feature of the contour sampling point, and the specific method of feature extraction, similarity calculation and key part identification is the same as the calculation of the x coordinate.

[0102] Embodiment 3:

[0103] A ship target key part identification system, comprising:

[0104] A detection module, used to obtain a radiometer image of a ship target;

[0105] The segmentation module is used to segment the ship target from the radiometer image, and to sample the contour curve or boundary of the ship target in the segmentation result to obtain a target contour sampling point set;

[0106] A feature extraction module, used to extract shape features of each contour sampling point in the target contour sampling point set;

[0107] A similarity calculation module is used to calculate the similarity between the shape features of each contour sampling point in the target contour sampling point set and the shape features of each contour sampling point of the template ship to obtain a similarity matrix; the template ship is a ship with known key parts;

[0108] And the matching module, for each key part of the template ship, takes the contour sampling points of the key part as the matching template, and selects the part of the contour sampling points with the highest similarity to each matching template from the target contour sampling point set according to the similarity matrix to locate the area where the corresponding key part of the ship target is located.

[0109] This embodiment is used to implement the method for identifying key parts of a ship target provided by the above-mentioned embodiment 1 or embodiment 2. The specific implementation methods of each module can refer to the description in the above-mentioned embodiment 1 or 2, and will not be repeated here.

[0110] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying key parts of a ship target, characterized in that: include: S1: Acquire the radiometer image of the ship target; S2: Segmenting a ship target from the radiometer image, and sampling a contour curve or a boundary of the ship target in the segmentation result to obtain a target contour sampling point set; S3: extracting shape features of each contour sampling point in the target contour sampling point set; S4: Calculate the similarity between the shape features of each contour sampling point in the target contour sampling point set and the shape features of each contour sampling point of the template ship to obtain a similarity matrix; the template ship is a ship with known key parts; S5: For each key part of the template ship, the contour sampling points of the key part are used as the matching template, and the partial contour sampling points with the highest similarity to each matching template are screened out from the target contour sampling points according to the similarity matrix to locate the area where the corresponding key part of the ship target is located.

2. The method for identifying key parts of a ship target according to claim 1, characterized in that: The shape feature of the contour sampling point is the x-coordinate or y-coordinate of the contour sampling point.

3. The method for identifying key parts of a ship target according to claim 2, characterized in that: In the step S4, a dynamic time warping DTW algorithm or a FastDTW algorithm is used to calculate the similarity between the shape features of each contour sampling point in the target contour sampling point set and the shape features of each contour sampling point of the template ship.

4. The method for identifying key parts of a ship target according to claim 3, characterized in that: In the step S2, full sampling is used to sample the contour curve or boundary of the ship target in the segmentation result.

5. The method for identifying key parts of a ship target according to claim 1, characterized in that: The shape features of the contour sampling points are shape context histogram features or directional gradient histogram features.

6. The method for identifying key parts of a ship target according to claim 5, characterized in that: In the step S4, the cosine similarity is used to calculate the similarity between the shape features of each contour sampling point in the target contour sampling point set and the shape features of each contour sampling point of the template ship.

7. The method for identifying key parts of a ship target according to claim 6, characterized in that: In the step S2, Jitendra sampling or uniform sampling is used to sample the contour curve or boundary of the ship target in the segmentation result.

8. The method for identifying key parts of a ship target according to any one of claims 2 to 7, characterized in that: Before extracting the shape features of each contour sampling point in the target contour sampling point set, the method further includes: correcting the orientation angle of the ship target so that the orientation angle of the ship target is the same as the orientation angle of the template ship.

9. The method for identifying key parts of a ship target according to claim 8, characterized in that: Correcting the orientation angle of the ship target includes: Performing ellipse fitting on the contour sampling points of the ship target; estimating the orientation angle of the ship target according to the ellipse fitting result; The orientation angle of the ship target is rotated to be the same as the orientation angle of the template ship.

10. A ship target key part identification system, characterized in that: include: A detection module, used to obtain a radiometer image of a ship target; A segmentation module, used to segment a ship target from the radiometer image, and sample the contour curve or boundary of the ship target in the segmentation result to obtain a target contour sampling point set; A feature extraction module, used to extract shape features of each contour sampling point in the target contour sampling point set; A similarity calculation module is used to calculate the similarity between the shape features of each contour sampling point in the target contour sampling point set and the shape features of each contour sampling point of the template ship to obtain a similarity matrix; the template ship is a ship with known key parts; And a matching module, for each key part of the template ship, takes the contour sampling points of the key part as the matching template, and selects some contour sampling points with the highest similarity to each matching template from the target contour sampling point set according to the similarity matrix to locate the area where the corresponding key part is located in the ship target.

Images