CN-116243268-B - Sea surface floating small target detection method and device based on feature optimization and false alarm controllable three-dimensional concave bag

CN116243268BCN 116243268 BCN116243268 BCN 116243268BCN-116243268-B

Abstract

The invention discloses a sea surface floating small target detection method and device based on feature optimization and false alarm controllable three-dimensional concave, comprising the steps of extracting sea clutter and eight-dimensional feature vectors of a sample to be detected from radar receiving echoes; the method comprises the steps of selecting optimal three-dimensional feature vectors from eight-dimensional feature vectors of sea clutter and samples to be detected respectively, taking the optimal three-dimensional feature vectors of the sea clutter as training sample points, taking the optimal three-dimensional feature vectors of the samples to be detected as the samples to be detected, using a false alarm controllable alpha bag algorithm, successively deleting false alarm points from the training sample points according to a false alarm rate on the basis of the principle of maximum bag area loss of single time to finish updating the training sample points, obtaining convex hull areas containing all the training sample points, converting the convex hull areas into bag judgment areas containing all the training sample points, and judging positions of the samples to be detected corresponding to the bag judgment areas to obtain detection results. On the premise of meeting the condition that the false alarm is controllable, accurate sea surface target detection is realized.

Inventors

SHI YANLING
HU YUEFENG
CHEN WEISHENG
GAO XINGYI

Assignees

南京邮电大学

Dates

Publication Date: 20260505
Application Date: 20230227

Claims (10)

1. The sea surface floating small target detection method based on feature optimization and false alarm controllable three-dimensional concave bag is characterized by comprising the following steps of: collecting radar receiving echoes, and extracting eight-dimensional feature vectors of sea clutter and eight-dimensional feature vectors of samples to be detected from the radar receiving echoes; Respectively selecting an optimal three-dimensional feature vector from eight-dimensional feature vectors of sea clutter and eight-dimensional feature vectors of samples to be detected by using a feature optimization algorithm, taking the optimal three-dimensional feature vector of the sea clutter as a training sample point, and taking the optimal three-dimensional feature vector of the samples to be detected as a sample point to be detected; The false alarm controllable alpha bag algorithm is used, the false alarm points are gradually deleted from the training sample points according to the false alarm rate on the principle that the single bag volume loss is maximum, and the updating of the training sample points is completed; obtaining a convex hull region containing all training sample points by using a convex hull learning algorithm, and converting the convex hull region into a concave hull judgment region containing all training sample points; and judging the position of the sample point to be detected corresponding to the concave packet judgment area to obtain a detection result.
2. The method for detecting the small floating target on the sea surface based on the feature preference and the false alarm controllable three-dimensional concave bag according to claim 1, wherein extracting the eight-dimensional feature vectors of the sea clutter and the sample to be detected comprises: normalized Hurst index NHE, relative average amplitude RAA, relative doppler peak height RDPH, relative vector entropy RVE, ridge accumulation RI, maximum connected region size MS, connected region number NR, generalized likelihood ratio detector NSCM-GLRT based on normalized sample covariance matrix.
3. The method for detecting the small floating target on the sea surface based on the feature optimization and the false alarm controllable three-dimensional concave bag according to claim 1, wherein the feature optimization algorithm is used for respectively selecting the optimal three-dimensional feature vector from eight-dimensional feature vectors of sea clutter and eight-dimensional feature vectors of samples to be detected, and the specific steps comprise: Splicing the eight-dimensional feature vector of the sea clutter and the eight-dimensional feature vector of the sample to be detected into a mixed eight-dimensional feature vector; selecting the best one-dimensional mixed feature vector from the mixed eight-dimensional feature vectors by utilizing a minimum redundancy maximum correlation feature optimization algorithm mRMR; Sequentially selecting a second best one-dimensional mixed feature vector and a third best one-dimensional mixed feature vector from the mixed eight-dimensional feature vectors by using mRMR, and forming an optimal three-dimensional mixed feature vector together with the optimal one-dimensional mixed feature vector; And separating the optimal three-dimensional mixed characteristic vector into a sea clutter optimal three-dimensional characteristic vector and a sample optimal three-dimensional characteristic vector to be detected.
4. The method for detecting the small sea surface floating target based on feature optimization and false alarm controllable three-dimensional concave bag according to any one of claims 1 to 3, wherein a false alarm controllable alpha concave bag algorithm is used, the false alarm points are gradually deleted from the training sample points according to the false alarm rate on the principle of maximum single concave bag area loss, and the updating of the training sample points is completed, and the specific steps comprise: step 1.1, calculating the number Q of sample points in a set zeta formed by training sample points; Step 1.2, calculating the number of false alarm points N f ＝Q×P F according to the false alarm rate P F ; Step 1.3, iteration times v=1, operation set ζ v =ζ; step 1.4, generating a concave packet about an operation set ζ v according to an alpha concave packet algorithm; step 1.5, deleting a concave vertex from the operation set zeta v , and calculating the concave volume; Step 1.6, searching a concave vertex which reduces the concave volume to the maximum; Step 1.7, deleting the concave vertex found in the step 1.6 from the operation set ζ v , wherein the iteration times v=v+1 to obtain a new operation set ζ v , returning to the step 1.4 if the iteration times v is less than or equal to N f , otherwise ending the iteration, and finally obtaining a set ζ Nf with the false alarm points deleted.
5. The method for detecting the small floating target on the sea surface based on feature preference and false alarm controllable three-dimensional concave bag according to claim 4, wherein the step of converting the convex bag area into the concave bag judgment area containing all training sample points comprises the following specific steps: Step 2.1, generating an original convex hull omega original with the surface consisting of D trilateral faces through a convex hull learning algorithm according to a set zeta Nf of the deleted false alarm points, wherein D is the number of the trilateral faces; Step 2.2, calculating the circumferences L d of all the trilateral faces of the original convex hull, d=0, 1, & D; Step 2.3, calculating the average value of the circumferences of all the trilateral faces, and taking the average value as a threshold value th; step 2.4, the number of endocutters i=1, the number of iterations j=1, the operation area omega i,j ＝Ω original is set up as the maximum endocutter number dig_num; Step 2.5, if j is less than or equal to dig_num, performing step 2.6-2.10, otherwise ending iteration, and jumping to step 2.11; step 2.6, calculating the circumferences L j,d of all the three-sided faces of the operation region Ω i,j , d=0, 1,. -%; step 2.7, calculating the circumferences L j,d of all the trilateral faces, d=0, 1..maximum value L j,max in D; Step 2.8, if L j,max > th is met, performing step 2.9-2.10, otherwise ending iteration, and jumping to step 2.11; Step 2.9, performing an endocutter algorithm operation on the operation area omega i,j , wherein the endocutter times i=i+1 and the iteration times j=j to obtain an updated operation area omega i,j ; Step 2.10, performing filling algorithm operation on the operation area omega i,j , wherein the iteration times j=j+1 and the endocut times i=i to obtain an updated operation area omega i,j , and jumping to step 2.5; And 2.11, obtaining a final concave packet judgment area omega final ＝Ω i,j .
6. The method for detecting a small target on the sea surface based on feature preference and false alarm controllable three-dimensional concave bag according to claim 5, wherein the step of operating the endocutter algorithm comprises: Step 3.1, inputting an operation area omega i,j ; Step 3.2, finding a first trilateral plane delta m1,i corresponding to the perimeter maximum L j,max ; Step 3.3, finding the longest side in the first trilateral plane delta m1,i to obtain a second trilateral plane delta m2,i sharing the side with the trilateral plane delta m1,i ; Step 3.4, removing the vertex related to the operation area omega i,j from the aggregate zeta Nf , taking the rest points as interior points, and finding out the point closest to the center point of the side from the interior points as an interior tangent point P 0 ; Step 3.5, setting all vertexes of the first trilateral plane delta m1,i and the second trilateral plane delta m2,i as a point set F, establishing a new plane delta 1,i ,△ 2,i ,△ 3,i ,△ 4,i by using the point set F and an inner tangent point P 0 , and satisfying the following formula: Step 3.6, deleting the first and second triangular surfaces delta m1,i ,△ m2,i from the operation area omega i,j , adding 4 surfaces delta 1,i ,△ 2,i ,△ 3,i ,△ 4,i to finish updating, and cutting off the hexahedral PO dig consisting of the 6 surfaces delta 1,i ,△ 2,i ,△ 3,i ,△ 4,i ,△ m1,i ,△ m2,i in the space corresponding to omega i,j at the space angle; And 3.7, outputting the updated operation area omega i,j .
7. The method for detecting a small target on the sea surface based on feature preference and false alarm controllable three-dimensional concave bag according to claim 6, wherein the step of filling algorithm operation comprises: Step 4.1, inputting an operation area omega i,j ; step 4.2, using training sample points Z h outside the operation region Ω i,j , composing a set z= { Z 1 ,...,Z H }, H being the number of Z h ; step 4.3, the first iteration times t=1 and the second iteration times h=1; Step 4.4, if t is less than or equal to 4, performing the steps 4.5-4.13, otherwise ending the iteration, and jumping to the step 4.14; step 4.5, calculating the distance from the external training sample point set Z to the face delta 1,i ,△ 2,i ,△ 3,i ,△ 4,i , namely a distance 1, a distance 2, a distance 3 and a distance 4:di 1 ,di 2 ,di 3 ,di 4 respectively; step 4.6, when the t-th iteration is performed, the distance from the external training sample point set Z to the face delta t,i is di t , if di t ＝min{di 1 ,di 2 ,di 3 ,di 4 is satisfied, step 4.7 is performed, and otherwise, step 4.5 is skipped; Step 4.7, storing the external training sample points Z h into a set J s , and jumping to step 4.5 if H is less than or equal to H and the second iteration times h=h+1 are met, otherwise jumping to step 4.8; Step 4.8, sorting the points in the set J s from near to far according to the distance plane delta t,i to obtain a new set J' s ＝{J' 1 ,...,J' S }, wherein S is the number of points in the set J s ; Step 4.9, third iteration number s=1; Step 4.10, finding the face delta near nearest to the point J' s in all the trilateral faces of the surface of the operation area omega i,j ; Step 4.11, setting all vertexes of the face delta near nearest to the point J ' s as a point set F'; Step 4.12, establishing a new face delta 5,j ,△ 6,j ,△ 7,j using point set F 'and point J' s , and satisfying: Step 4.13, deleting the face delta near nearest to the point J' s from the operation area omega i,j , adding the 3 faces delta 5,j ,△ 6,j ,△ 7,j to finish updating omega i,j , filling tetrahedron PO fill consisting of the 4 faces delta 5,j ,△ 6,j ,△ 7,j ,△ near in the space angle equivalent to omega i,j , and the third iteration number s=s+1, wherein the first iteration number t=t+1, if S is less than or equal to S, jumping to step 4.10, otherwise jumping to step 4.4; And 4.14, outputting the updated operation area omega i,j .
8. The method for detecting a small sea surface floating target based on feature preference and false alarm controllable three-dimensional concave bag according to claim 7, wherein the step of judging the position of the sample point to be detected corresponding to the concave bag judgment area to obtain the detection result comprises the following steps: Firstly judging whether a sample point to be detected is in an original convex hull omega original or not, if not, judging that the sample to be detected is a target, otherwise, continuously judging whether the sample point to be detected is in a hexahedral PO dig or not, if not, judging that the sample to be detected is sea clutter, otherwise, continuously judging whether the sample point to be detected is in a tetrahedron PO fill or not, if not, judging that the sample to be detected is the target, and otherwise, judging that the sample to be detected is the sea clutter.
9. Sea surface floating small target detection device based on feature optimization and false alarm controllable three-dimensional concave bag, and is characterized by comprising: The data acquisition module is used for acquiring radar receiving echoes, and extracting eight-dimensional feature vectors of sea clutter and eight-dimensional feature vectors of samples to be detected from the radar receiving echoes; The data processing module performs the following operations: Respectively selecting an optimal three-dimensional feature vector from eight-dimensional feature vectors of sea clutter and eight-dimensional feature vectors of samples to be detected by using a feature optimization algorithm, taking the optimal three-dimensional feature vector of the sea clutter as a training sample point, and taking the optimal three-dimensional feature vector of the samples to be detected as a sample point to be detected; The training module performs the following operations: The false alarm controllable alpha bag algorithm is used, the false alarm points are gradually deleted from the training sample points according to the false alarm rate on the principle that the single bag volume loss is maximum, and the updating of the training sample points is completed; obtaining a convex hull region containing all training sample points by using a convex hull learning algorithm, and converting the convex hull region into a concave hull judgment region containing all training sample points; and the detection module is used for judging the position of the sample point to be detected corresponding to the concave packet judgment area to obtain a detection result.
10. The sea surface floating small target detection device based on feature optimization and false alarm controllable three-dimensional concave bag according to claim 9, wherein the training module uses a false alarm controllable alpha concave bag algorithm, and based on the principle of maximum single concave bag area loss, successively deletes false alarm points from training sample points according to the false alarm rate, and completes updating the training sample points, and the specific steps include: step 1.1, calculating the number Q of sample points in a set zeta formed by training sample points; Step 1.2, calculating the number of false alarm points N f ＝Q×P F according to the false alarm rate P F ; Step 1.3, iteration times v=1, operation set ζ v =ζ; step 1.4, generating a concave packet about an operation set ζ v according to an alpha concave packet algorithm; step 1.5, deleting a concave vertex from the operation set zeta v , and calculating the concave volume; Step 1.6, searching a concave vertex which reduces the concave volume to the maximum; step 1.7, deleting the concave vertex found in the step 1.6 from the operation set ζ v , wherein the iteration times v=v+1 to obtain a new operation set ζ v , returning to the step 1.4 if the iteration times v is less than or equal to N f , otherwise ending the iteration, and finally obtaining a set ζ Nf with the false alarm points deleted; The convex hull area is converted into a concave hull judgment area containing all training sample points, and the specific steps comprise: Step 2.1, generating an original convex hull omega original with the surface consisting of D trilateral faces through a convex hull learning algorithm according to a set zeta Nf of the deleted false alarm points, wherein D is the number of the trilateral faces; Step 2.2, calculating the circumferences L d of all the trilateral faces of the original convex hull, d=0, 1, & D; Step 2.3, calculating the average value of the circumferences of all the trilateral faces, and taking the average value as a threshold value th; step 2.4, the number of endocutters i=1, the number of iterations j=1, the operation area omega i,j ＝Ω original is set up as the maximum endocutter number dig_num; Step 2.5, if j is less than or equal to dig_num, performing step 2.6-2.10, otherwise ending iteration, and jumping to step 2.11; step 2.6, calculating the circumferences L j,d of all the three-sided faces of the operation region Ω i,j , d=0, 1,. -%; step 2.7, calculating the circumferences L j,d of all the trilateral faces, d=0, 1..maximum value L j,max in D; Step 2.8, if L j,max > th is met, performing step 2.9-2.10, otherwise ending iteration, and jumping to step 2.11; Step 2.9, performing an endocutter algorithm operation on the operation area omega i,j , wherein the endocutter times i=i+1 and the iteration times j=j to obtain an updated operation area omega i,j ; Step 2.10, performing filling algorithm operation on the operation area omega i,j , wherein the iteration times j=j+1 and the endocut times i=i to obtain an updated operation area omega i,j , and jumping to step 2.5; step 2.11, obtaining a final concave packet judgment area omega final ＝Ω i,j ; The operation steps of the endocarp algorithm comprise: Step 3.1, inputting an operation area omega i,j ; Step 3.2, finding a first trilateral plane delta m1,i corresponding to the perimeter maximum L j,max ; Step 3.3, finding the longest side in the first trilateral plane delta m1,i to obtain a second trilateral plane delta m2,i sharing the side with the trilateral plane delta m1,i ; Step 3.4, removing the vertex related to the operation area omega i,j from the aggregate zeta Nf , taking the rest points as interior points, and finding out the point closest to the center point of the side from the interior points as an interior tangent point P 0 ; Step 3.5, setting all vertexes of the first trilateral plane delta m1,i and the second trilateral plane delta m2,i as a point set F, establishing a new plane delta 1,i ,△ 2,i ,△ 3,i ,△ 4,i by using the point set F and an inner tangent point P 0 , and satisfying the following formula: Step 3.6, deleting the first and second triangular surfaces delta m1,i ,△ m2,i from the operation area omega i,j , adding 4 surfaces delta 1,i ,△ 2,i ,△ 3,i ,△ 4,i to finish updating, and cutting off the hexahedral PO dig consisting of the 6 surfaces delta 1,i ,△ 2,i ,△ 3,i ,△ 4,i ,△ m1,i ,△ m2,i in the space corresponding to omega i,j at the space angle; Step 3.7, outputting the updated operation area omega i,j ; the steps of the filling algorithm operation include: Step 4.1, inputting an operation area omega i,j ; step 4.2, using training sample points Z h outside the operation region Ω i,j , composing a set z= { Z 1 ,...,Z H }, H being the number of Z h ; step 4.3, the first iteration times t=1 and the second iteration times h=1; Step 4.4, if t is less than or equal to 4, performing the steps 4.5-4.13, otherwise ending the iteration, and jumping to the step 4.14; step 4.5, calculating the distance from the external training sample point set Z to the face delta 1,i ,△ 2,i ,△ 3,i ,△ 4,i , namely a distance 1, a distance 2, a distance 3 and a distance 4:di 1 ,di 2 ,di 3 ,di 4 respectively; step 4.6, when the t-th iteration is performed, the distance from the external training sample point set Z to the face delta t,i is di t , if di t ＝min{di 1 ,di 2 ,di 3 ,di 4 is satisfied, step 4.7 is performed, and otherwise, step 4.5 is skipped; Step 4.7, storing the external training sample points Z h into a set J s , and jumping to step 4.5 if H is less than or equal to H and the second iteration times h=h+1 are met, otherwise jumping to step 4.8; Step 4.8, sorting the points in the set J s from near to far according to the distance plane delta t,i to obtain a new set J' s ＝{J' 1 ,...,J' S }, wherein S is the number of points in the set J s ; Step 4.9, third iteration number s=1; Step 4.10, finding the face delta near nearest to the point J' s in all the trilateral faces of the surface of the operation area omega i,j ; Step 4.11, setting all vertexes of the face delta near nearest to the point J ' s as a point set F'; Step 4.12, establishing a new face delta 5,j ,△ 6,j ,△ 7,j using point set F 'and point J' s , and satisfying: Step 4.13, deleting the face delta near nearest to the point J' s from the operation area omega i,j , adding the 3 faces delta 5,j ,△ 6,j ,△ 7,j to finish updating omega i,j , filling tetrahedron PO fill consisting of the 4 faces delta 5,j ,△ 6,j ,△ 7,j ,△ near in the space angle equivalent to omega i,j , and the third iteration number s=s+1, wherein the first iteration number t=t+1, if S is less than or equal to S, jumping to step 4.10, otherwise jumping to step 4.4; step 4.14, outputting the updated operation area omega i,j ; the step of judging the position of the sample point to be detected corresponding to the concave packet judgment area to obtain a detection result comprises the following steps: Firstly judging whether a sample point to be detected is in an original convex hull omega original or not, if not, judging that the sample to be detected is a target, otherwise, continuously judging whether the sample point to be detected is in a hexahedral PO dig or not, if not, judging that the sample to be detected is sea clutter, otherwise, continuously judging whether the sample point to be detected is in a tetrahedron PO fill or not, if not, judging that the sample to be detected is the target, and otherwise, judging that the sample to be detected is the sea clutter.

Description

Sea surface floating small target detection method and device based on feature optimization and false alarm controllable three-dimensional concave bag Technical Field The invention belongs to the field of radar target classification, and particularly relates to a sea surface floating small target detection method and device based on feature optimization and false alarm controllable three-dimensional concave bag. Background Under the background of sea clutter, the detection of a small floating target on the sea surface is always the key point and the difficulty of research of domestic and foreign experts and scholars. Since small targets have a small radar cross-sectional area (Radar Cross Section, RCS) and weak radar echoes, the problem of low detection probability of conventional energy-based detectors is often caused, and non-energy feature detection technology is an effective way to solve the problem. Early feature detectors performed from a single feature point of view, distinguishing between sea clutter and target echoes by studying their different characteristics over numerous transform domains. However, the single-characteristic target detection algorithm has certain limitations in coping with different sea conditions. In order to use more features to improve detection performance, shui et al propose a three-feature convex hull detector based on amplitude and doppler spectrum, which realizes joint detection of time domain and frequency domain features. Because the sea clutter and the target echo have stronger separability in the time-frequency three-feature space, scholars such as Shui and the like propose a convex hull detector based on the time-frequency three-feature on the basis, and the performance of the detector is greatly improved. However, by comparing the detection results, it is found that the two three-feature convex hull detection algorithms are excellent in different data. In addition, the distribution of the sea clutter in the three-dimensional feature space depends on the selection of the features, and in most cases, the distribution is concave, and when a convex hull is utilized to determine a region to be detected formed by the features of the sea clutter, the concave distribution is forcedly changed into the convex distribution, so that a judgment region is necessarily enlarged, and the detection probability is further lost. Furthermore, in the big data age, both the data volume and the data dimension are increasing. How to extract features that are highly diverse in high-dimensional features is a critical issue. Feature selection is one of the currently available dimension reduction techniques that reduces the dimension of data by removing correlated and redundant features while preserving uncorrelated features to form an optimal feature subset. Disclosure of Invention Aiming at the problem that sea clutter is affected when a small sea surface floating target is detected, the invention provides a sea surface small floating target detection method and a device based on feature optimization and false alarm controllable three-dimensional concave, wherein a mRMR algorithm is used for removing feature vectors with high redundancy and correlation, an optimal three-dimensional feature vector is selected, and the detection is completed by combining the false alarm controllable three-dimensional concave algorithm, so that excellent detection performance is obtained. The technical scheme adopted by the invention is as follows: In a first aspect, the invention provides a sea surface floating small target detection method based on feature optimization and false alarm controllable three-dimensional concave, which comprises the following steps: collecting radar receiving echoes, and extracting eight-dimensional feature vectors of sea clutter and eight-dimensional feature vectors of samples to be detected from the radar receiving echoes; Respectively selecting an optimal three-dimensional feature vector from eight-dimensional feature vectors of sea clutter and eight-dimensional feature vectors of samples to be detected by using a feature optimization algorithm, taking the optimal three-dimensional feature vector of the sea clutter as a training sample point, and taking the optimal three-dimensional feature vector of the samples to be detected as a sample point to be detected; The false alarm controllable alpha bag algorithm is used, the false alarm points are gradually deleted from the training sample points according to the false alarm rate on the principle that the single bag volume loss is maximum, and the updating of the training sample points is completed; obtaining a convex hull region containing all training sample points by using a convex hull learning algorithm, and converting the convex hull region into a concave hull judgment region containing all training sample points; and judging the position of the sample point to be detected corresponding to the concave packet judgment area to ob