CN111707998A - Sea surface floating small target detection method based on connected region characteristics - Google Patents

Sea surface floating small target detection method based on connected region characteristics Download PDF

Info

Publication number
CN111707998A
CN111707998A CN202010543795.6A CN202010543795A CN111707998A CN 111707998 A CN111707998 A CN 111707998A CN 202010543795 A CN202010543795 A CN 202010543795A CN 111707998 A CN111707998 A CN 111707998A
Authority
CN
China
Prior art keywords
time
training
frequency
unit
target
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010543795.6A
Other languages
Chinese (zh)
Other versions
CN111707998B (en
Inventor
许述文
陈康权
白晓惠
水鹏朗
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xidian University
Original Assignee
Xidian University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xidian University filed Critical Xidian University
Priority to CN202010543795.6A priority Critical patent/CN111707998B/en
Publication of CN111707998A publication Critical patent/CN111707998A/en
Application granted granted Critical
Publication of CN111707998B publication Critical patent/CN111707998B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
    • G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
    • G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
    • G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
    • G01S7/414—Discriminating targets with respect to background clutter
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Abstract

The invention discloses a method for detecting small sea surface floating targets based on characteristics of a connected region, which mainly solves the problem that the probability of detecting the small sea surface targets in a short observation time is low in the prior art; the realization process is as follows: 1) constructing a U-Net network; 2) selecting radar echo data of a training unit containing a target to perform block whitening, and manufacturing a time-frequency graph training set; 3) training the U-Net network by using a time-frequency diagram data set of a training unit; 4) selecting reference unit echo data containing pure clutter to perform block whitening, making a time-frequency diagram set, inputting the time-frequency diagram set into a trained U-Net network to obtain an extraction result of a connected region of the time-frequency diagram, performing binarization processing on the extraction result, and calculating the area of the connected region to obtain a test statistic D; 5) obtaining a detection threshold T through Monte Carlo simulation; 6) and judging whether a target exists according to a comparison result of the detection statistic D and the detection threshold T, judging that the target exists if the detection statistic D is larger than or equal to the detection threshold T, and otherwise, judging that the target does not exist.

Description

Sea surface floating small target detection method based on connected region characteristics
Technical Field
The invention relates to the technical field of signal processing, in particular to a method for detecting a small floating target on the sea surface based on the characteristics of a connected region.
Background
The sea clutter is radar echo received by a radar and reflected from the sea surface, and when the sea surface search radar detects the sea surface, the sea clutter inevitably influences the detection of floating small targets such as floating ice, boats, navigation marks and the like on the sea surface. The intensity of the sea clutter varies with the radar parameters, the direction of illumination, the sea state, etc. Due to the space-time non-stationarity of the high-resolution sea clutter, the traditional target detection method has the problems of low detection probability and high false alarm, so that the detection of the small targets floating on the sea surface under the background of the sea clutter becomes a difficult point.
In response to this problem, many scholars have made extensive attempts and studies. The adaptive matched filtering method is proposed on the premise that sea clutter satisfies some statistical models. And obtaining the optimal K distribution detector under the K distribution sea clutter model. When the sea clutter sequence satisfies the Pareto distribution, an optimal detector under the sea clutter model can be obtained. However, because the existing statistical model is difficult to describe the complex characteristics of the sea clutter, the sea clutter has a wide doppler bandwidth, the sea surface target is usually low in speed, and the target is usually submerged in the sea clutter bandwidth, the performance of the detector is greatly degraded when the method detects the sea surface target, particularly a floating small target.
In addition, the fractal-based target detection method can achieve a good detection result when the observation time is long, but when the observation time is short, the detection performance is obviously reduced, and the requirement on the sea search radar cannot be met.
Although, detection methods based on time-frequency analysis have received attention from many scholars. However, the sea clutter is a result of superposition of backscattering vectors of a large number of scatterers with different speeds, the sea clutter after the block whitening processing becomes approximate white noise, the clutter consists of small communicated areas which are broken, discontinuous, irregular and discrete in distribution on a time-frequency plane, and the energy is low. In contrast, under the action of inertia, the radial velocities of strong scattering points on the targets relative to the radar in a short time can be regarded as approximately the same, so that the target energy is concentrated mainly on a curve close to a straight line near 0 Hz. The target appears as a more concentrated connected region on the time-frequency plane, and the energy is higher. How to extract features for detecting a target from the difference of connected regions of the target and the clutter time-frequency map is the focus of target detection. The problem of feature extraction based on the time-frequency image connected region can be further refined to the problem of image segmentation. In a traditional image segmentation algorithm, such as a threshold segmentation algorithm, a gray level histogram of an image is divided into classes by one or more thresholds, and pixels with gray levels in the same class in the image are considered to belong to the same object. Since the gradation characteristic of the image is directly used, the calculation is simple. However, this method only considers the gray level of the pixel itself, and generally does not consider the spatial characteristics, so it is sensitive to noise, and the selection of the threshold is also a key and difficult point. For example, in an algorithm based on mathematical morphology, a probe called a structural element is used to collect information of an image, and the structural features of the image are known through continuous movement of the probe on the image, so that the positioning effect is good. However, after image processing, there are still a lot of short lines and isolated points that do not match the target, and the selection of structural elements also directly affects the segmentation effect.
Disclosure of Invention
The invention aims to provide a method for detecting small sea surface floating targets based on characteristics of a connected region, which can achieve the effect of distinguishing sea clutter from targets by intelligently extracting the characteristics of the connected region through a U-Net neural network, thereby improving the detection performance of the small sea surface floating targets and meeting the requirement of a radar on sea search.
The technical scheme adopted by the invention is as follows:
step 1) constructing a U-Net network
The U-Net network is formed by repeatedly splicing a convolutional layer, a pooling layer and a regularization layer, the convolutional layer output is processed by using a ReLU activation function, and the size of a convolutional kernel is set to be 3 × 3 × nfAnd the number n of downsampling kernels at a timefDoubling, the number of cores to be sampled n at each timefHalving;
step 2) making a U-Net network training data set
Step 2a) acquiring radar echo data:
transmitting signals to the sea surface by means of a radar transmitterThe radar receiver receives echo data reflected by the sea surface to acquire radar echo data, and the echo data are divided into pure clutter data and echo data containing targets; selecting a part of distance units from the echo data containing the target as training units, wherein the time sequence of the training units is as follows: z (N), N ═ 1, 2.., N; taking the distance unit of the pure clutter as a reference unit, wherein the time sequence of the reference unit is as follows: z is a radical ofp(N), N1, 2, N, p 1,2, Q is the number of reference units, N is the length of the time series;
step 2b) of comparing the training unit time series z with the reference unit time series zpCarrying out block whitening, and then iterating by using an approximate maximum likelihood estimation method to obtain a speckle covariance matrix of a clutter vector of each subinterval; carrying out block whitening on the training unit echo and the reference unit echo by using the iterated covariance matrix to obtain a whitened training unit echo vector and a whitened reference unit echo vector;
step 2c), making a whitened time-frequency diagram data set:
calculating a smooth Wigner-Willi distribution of the training units and a smooth Wigner-Willi distribution of the reference units; then storing the smooth Wigner-Willi distribution of the training unit and the smooth Wigner-Willi distribution of the reference unit as a time-frequency image format to obtain a whitened training unit time-frequency image data set and a whitened reference unit time-frequency image data set;
step 2d), manufacturing a training label corresponding to the whitening time-frequency diagram data set of the training unit:
using an image labeling tool labelme to label a connected region where a target in a training unit time-frequency image data set is located, generating a json file after labeling, and analyzing the json file to obtain a required label image set;
step 3), training the U-Net network:
adopting the time-frequency graphs of 300 training units as a training set, adopting the time-frequency graphs of 30 training units as a verification set, and carrying out network training by using an Adam algorithm;
step 4), calculating detection statistics:
inputting the time-frequency graph into a U-Net network to obtain a connected region extraction graph in an RGB format, converting the connected region extraction graph into a binary image, setting a background gray value to be 0 and a target gray value to be 255, and obtaining the area of the connected region by counting the number of pixel points greater than 0, namely detecting statistic D;
step 5), determining a detection threshold:
inputting a whitened time-frequency diagram data set of a reference unit containing pure clutter into a trained U-Net network, calculating a detection statistic D, and obtaining a detection threshold T through Monte Carlo simulation;
step 6), detection:
inputting the change time-frequency diagram of the unit to be detected into the trained U-Net network, and calculating a detection statistic D; wherein the unit to be detected is an unknown unit; comparing the detection statistic D with the detection threshold T, and judging the target detection according to the comparison result:
if D is larger than or equal to T, a target is considered;
if D < T, no target is considered.
The specific steps of the step 2b) are as follows:
training unit time sequence z and reference unit time sequence zpRespectively cut into short vectors with length M and no overlap, namely:
z＝[z1,z2,...,zm,...,zN/M]T
zp＝[zp,1,zp,2,...,zp,m,...,zp,N/M]T,p＝1,2,...,Q
wherein the vector zmM short vectors, z, representing training unit time trainingp,mAn mth short vector representing a reference unit time sequence, wherein M is 1,2, and N/M, and a speckle covariance matrix of a clutter vector of each subinterval is obtained by iteration through an approximate maximum likelihood estimation method:
Figure BDA0002539947560000041
Figure BDA0002539947560000042
and (3) carrying out block whitening on the training unit echo and the reference unit echo by using the covariance matrix after iteration to obtain a whitened training unit echo vector and a whitened reference unit echo vector:
Figure BDA0002539947560000043
Figure BDA0002539947560000044
the specific steps of the step 2c) are as follows:
step 2c), making a whitened time-frequency diagram data set:
calculating a smooth Wigner-Willi distribution of the training units and a smooth Wigner-Willi distribution of the reference units:
Figure RE-GDA0002609681030000045
Figure RE-GDA0002609681030000046
wherein the superscript denotes the conjugate, g (m) is the time smoothing window, h (k) is the frequency smoothing window, E denotes half the time smoothing window length, F denotes half the frequency smoothing window length, Δ FdA sampling interval that is a normalized doppler frequency, N1, 2.. and N, l 1, 2.. and N;
and storing the time-frequency matrixes of the training units calculated by the smooth Wigner-Willi distribution and the smooth Wigner-Willi distribution of the reference units into an image format to obtain a whitened training unit time-frequency image data set.
Compared with the prior art, the invention has the following advantages:
1) according to the invention, the time-frequency characteristics are automatically extracted from the time-frequency graphs of the sea clutter and the targets by using U-Net, and the detection of the small floating targets on the sea surface is completed by using the difference of the geometrical characteristics of the connected region of the time-frequency graphs of the sea clutter and the targets, so that a better detection effect can be obtained in a shorter observation time compared with a fractal-based single-feature detector;
2) the method utilizes U-Net to realize the extraction of the connected region of the time-frequency image, and compared with the traditional image segmentation algorithm, the method can better solve the problems of noise and nonuniformity in the image, reduce human intervention and obtain better segmentation effect;
3) the U-Net utilized by the invention adopts the same-layer resolution cascade between the down sampling and the up sampling in the structure, so that the low-layer characteristic information and the high-layer characteristic information are combined to be learned together, the network segmentation effect is more accurate, the generalization capability of the network is stronger, and a better effect can be achieved on a training set with a small sample. The problem that small target samples floating on the sea surface are scarce is solved to a certain extent. Compared with the traditional neural network, the requirement for a large amount of training data is reduced.
Drawings
FIG. 1 is a flow chart of an implementation of the present invention;
FIG. 2 is a U-Net network structure designed specifically in the present invention;
FIG. 3 is a result of feature extraction of connected regions of sea clutter data;
FIG. 4 is a connected component feature extraction result of target data;
FIG. 5 is a graph a comparing the detection performance of the present invention and the existing fractal feature based detection method under four polarizations when the observation time is 512 ms;
FIG. 6 is a graph b comparing the detection performance of the present invention and the existing fractal feature-based detection method under four polarizations when the observation time is 512 ms;
FIG. 7 is a graph c comparing the detection performance of the present invention and the existing fractal feature based detection method under four polarizations when the observation time is 512 ms;
fig. 8 is a graph d comparing the detection performance of the present invention with that of the existing fractal feature-based detection method under four polarizations when the observation time is 512 ms.
Detailed Description
As shown in fig. 1, the present invention comprises the steps of:
step 1) constructing a U-Net network
The V-Net network is formed by repeatedly splicing a convolutional layer, a pooling layer and a regularization layer, the convolutional layer output is processed by using a ReLU activation function, and the size of a convolutional kernel is set to be 3 × 3 × nfAnd the number n of downsampling kernels at a timefDoubling, the number of cores to be sampled n at each timefHalving;
step 2) making a U-Net network training data set
Step 2a) acquiring radar echo data:
sending a signal to the sea surface by using a radar transmitter, receiving echo data reflected by the sea surface by using a radar receiver, and acquiring radar echo data, wherein the echo data is divided into pure clutter data and echo data containing a target; selecting a part of distance units from the echo data containing the target as training units, wherein the time sequence of the training units is as follows: z (N), N ═ 1, 2.., N; taking the distance unit of the pure clutter as a reference unit, wherein the time sequence of the reference unit is as follows: z is a radical ofp(N), N1, 2, N, p 1,2, Q is the number of reference units, N is the length of the time series;
step 2b) of comparing the training unit time series z with the reference unit time series zpCarrying out block whitening, and then iterating by using an approximate maximum likelihood estimation method to obtain a speckle covariance matrix of a clutter vector of each subinterval; carrying out block whitening on the training unit echo and the reference unit echo by using the iterated covariance matrix to obtain a whitened training unit echo vector and a whitened reference unit echo vector;
step 2c), making a whitened time-frequency diagram data set:
calculating a smooth Wigner-Willi distribution of the training units and a smooth Wigner-Willi distribution of the reference units; then storing the smooth Wigner-Willi distribution of the training unit and the smooth Wigner-Willi distribution of the reference unit as a time-frequency image format to obtain a whitened training unit time-frequency image data set and a whitened reference unit time-frequency image data set;
step 2d), manufacturing a training label corresponding to the whitening time-frequency diagram data set of the training unit:
using an image labeling tool labelme to label a connected region where a target in a training unit time-frequency image data set is located, generating a json file after labeling, and analyzing the json file to obtain a required label image set;
step 3), training the U-Net network:
adopting the time-frequency graphs of 300 training units as a training set, adopting the time-frequency graphs of 30 training units as a verification set, and carrying out network training by using an Adam algorithm;
step 4), calculating detection statistics:
inputting the time-frequency graph into a U-Net network to obtain a connected region extraction graph in an RGB format, converting the connected region extraction graph into a binary image, setting a background gray value to be 0 and a target gray value to be 255, and obtaining the area of the connected region by counting the number of pixel points greater than 0, namely detecting statistic D;
step 5), determining a detection threshold:
inputting a whitened time-frequency diagram data set of a reference unit containing pure clutter into a trained U-Net network, calculating a detection statistic D, and obtaining a detection threshold T through Monte Carlo simulation;
step 6), detection:
inputting the change time-frequency diagram of the unit to be detected into the trained U-Net network, and calculating a detection statistic D; wherein the unit to be detected is an unknown unit; comparing the detection statistic D with the detection threshold T, and judging the target detection according to the comparison result:
if D is larger than or equal to T, a target is considered;
if D < T, no target is considered.
The specific steps of the step 2b) are as follows:
training unit time sequence z and reference unit time sequence zpRespectively cut into short vectors with length M and no overlap, namely:
z＝[z1,z2,...,zm,...,zN/M]T
zp＝[zp,1,zp,2,...,zp,m,...,zp,N/M]T,p＝1,2,...,Q
wherein the vector zmM short vectors, z, representing training unit time trainingp,mAn mth short vector representing a reference unit time sequence, wherein M is 1,2, and N/M, and a speckle covariance matrix of a clutter vector of each subinterval is obtained by iteration through an approximate maximum likelihood estimation method:
Figure BDA0002539947560000071
Figure BDA0002539947560000072
and (3) carrying out block whitening on the training unit echo and the reference unit echo by using the covariance matrix after iteration to obtain a whitened training unit echo vector and a whitened reference unit echo vector:
Figure BDA0002539947560000073
Figure BDA0002539947560000074
the specific steps of the step 2c) are as follows:
step 2c), making a whitened time-frequency diagram data set:
calculating a smooth Wigner-Willi distribution of the training units and a smooth Wigner-Willi distribution of the reference units:
Figure RE-GDA0002609681030000075
Figure RE-GDA0002609681030000076
wherein the superscript denotes the conjugate, g (m) is the time smoothing window, h (k) is the frequency smoothing window, E denotes half the time smoothing window length, F denotes half the frequency smoothing window length, Δ FdTo normalizeSampling interval of doppler frequency, N ═ 1, 2., N, l ═ 1, 2., N;
and storing the time-frequency matrixes of the training units calculated by the smooth Wigner-Willi distribution and the smooth Wigner-Willi distribution of the reference units into an image format to obtain a whitened training unit time-frequency image data set.
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention.
Referring to fig. 1, the present invention includes two parts of training and detecting, and the specific steps are as follows:
step 1) constructing a U-Net network according to the figure 2, wherein the U-Net network is formed by repeatedly splicing a convolutional layer, a pooling layer and a regularization layer, the convolutional layer output is processed by using a ReLU activation function, and the size of a convolutional core is set to be 3 × 3 × nfAnd the number n of downsampling kernels at a timefDoubling the number of cores n sampled at a timefAnd (4) halving. The initial value of the number of cores in this example is selected as: n isf0＝16；
Step 2) making a U-Net network training data set
Step 2a) acquiring radar echo data:
sending a signal to the sea surface by using a radar transmitter, receiving echo data reflected by the sea surface by using a radar receiver, and acquiring radar echo data, wherein the echo data is divided into pure clutter data and echo data containing a target; selecting a part of distance units from the echo data containing the target as training units, wherein the time sequence of the training units is as follows: z (N), N ═ 1, 2.., N; taking the distance unit of the pure clutter as a reference unit, wherein the time sequence of the reference unit is as follows: z is a radical ofp(N), N1, 2, N, p 1,2, Q is the number of reference units, N is the length of the time series;
step 2b) of comparing the training unit time series z with the reference unit time series zpBlock whitening is carried out:
training unit time sequence z and reference unit time sequence zpRespectively cut into short vectors with non-overlapping lengths M, namely:
z＝[z1,z2,...,zm,...,zN/M]T
zp＝[zp,1,zp,2,...,zp,m,...,zp,N/M]T,p＝1,2,...,Q
wherein the vector zmM short vectors, z, representing training unit time trainingp,mAn mth short vector representing a reference unit time sequence, M1, 2.
And (3) iteratively obtaining a speckle covariance matrix of the clutter vector of each subinterval by using an approximate maximum likelihood estimation method:
Figure BDA0002539947560000091
Figure BDA0002539947560000092
the covariance matrix converges with a few iterations. In this example, i is 3. And (3) carrying out block whitening on the training unit echo and the reference unit echo by using the covariance matrix after iteration to obtain a whitened training unit echo vector and a whitened reference unit echo vector:
Figure BDA0002539947560000093
Figure BDA0002539947560000094
step 2c), making a whitened time-frequency diagram data set:
calculating a smooth Wigner-Willi distribution of the training units and a smooth Wigner-Willi distribution of the reference units:
Figure RE-GDA0002609681030000095
Figure RE-GDA0002609681030000096
wherein the superscript denotes the conjugate, g (m) is the time smoothing window, h (k) is the frequency smoothing window, E denotes half the time smoothing window length, F denotes half the frequency smoothing window length, Δ FdA sampling interval that is a normalized doppler frequency, N1, 2.. and N, l 1, 2.. and N;
the time smoothing window and the frequency smoothing window are not limited to Hanning window, Hamming window, Blackman window, and Keseph window, but in this example, the Keseph window with length of 31 is used as the time smoothing window, and the Keseph window with length of 63 is used as the frequency smoothing window;
storing the time-frequency matrixes of the training units calculated by the smooth Virger-Weili distribution and the smooth Virger-Weili distribution of the reference units into an image format to obtain a whitened training unit time-frequency image data set;
step 2d), manufacturing a training label corresponding to the whitening time-frequency diagram data set of the training unit:
and marking a connected region where the target is located in the training unit time-frequency graph data set by using an image marking tool labelme, generating a json file after marking, and analyzing the json file to obtain a required label picture set.
Step 3), training the U-Net network:
due to the structural characteristics of the U-Net and the geometric characteristics of the time-frequency diagram data set, the network can achieve good performance by adopting the time-frequency diagram data set training of the small sample training unit. In the example, the time-frequency diagrams of 300 training units are used as a training set, and the time-frequency diagrams of 30 training units are used as a verification set to perform network training. The optimization algorithm adopted by the embodiment is an Adam algorithm, but is not limited to an SGD algorithm, an Adagarad algorithm and a RMSProp algorithm, and because the SGD algorithm takes longer time than other algorithms, the Adam algorithm has the advantages of both the Adagarad algorithm and the RMSProp algorithm, and the Adam algorithm is usually the best effect;
step 4), calculating detection statistics:
inputting the time-frequency graph into a U-Net network to obtain a connected region extraction graph in an RGB format, converting the extracted graph into a binary image, setting a background gray value to be 0 and a target gray value to be 255, and counting the number of pixel points larger than 0 to obtain the area of a connected region, namely a detection statistic D;
step 5), determining a detection threshold:
and inputting the whitened time-frequency diagram data set of the reference unit containing the pure clutter into a trained U-Net network, calculating a detection statistic D, and obtaining a detection threshold T through Monte Carlo simulation. In the embodiment, 10000 pure clutter time-frequency graphs are adopted to determine a detection threshold T;
step 6), detection:
and inputting the change time-frequency diagram of the unit to be detected into the trained U-Net network, and calculating the detection statistic D. Wherein the unit to be detected is an unknown unit; in actual detection, whether the unit to be detected is a training unit containing a target or a reference unit only containing sea clutter cannot be determined, and such a unit is referred to as a unit to be detected. Comparing the detection statistic D with the detection threshold T, and judging the target detection according to the comparison result:
if D is larger than or equal to T, a target is considered;
if D < T, no target is considered.
The effect of the present invention will be further explained with the simulation experiment.
1. Experimental data
The actual measurement sea clutter data used in this example are 12 sets. Each set of data includes four kinds of polarization data, two kinds of which are homopolar data HH, VV and two kinds of which are cross-polarization data HV, VH. Of these, 10 sets of IPIX radar data from 1993, the target to be measured was a floating sphere 1m in diameter wrapped with aluminum wire, each polarization data contained 14 distance units, and the data length was 217(ii) a The remaining 2 sets of IPIX radar data from 1998, with the target to be tested being a floating boat, each polarization data consisting of 28 range cells, and a data length of 60000.
2. Simulation experiment
Fig. 3 and 4 show the extraction effect of the trained U-Net on the connected region of the target time-frequency diagram and the clutter time-frequency diagram respectively. Therefore, the extracted characteristics of the connected region have larger difference for the target time-frequency diagram with concentrated energy and the clutter time-frequency diagram with relatively dispersed energy. When the observation time is 512ms, the radar detection performance is simulated and compared under four polarization data by using the method and the fractal-based detection method, and the result is shown in fig. 5-8. Wherein FIG. 5(a) is a comparison graph of radar detection performance under the same-direction HH polarization data; FIG. 6(b) is a comparison graph of radar detection performance under equidirectional VV polarization data; FIG. 7(c) is a comparison graph of radar detection performance under heterodromous HV polarization data; fig. 8(d) is a radar detection performance comparison chart under the anisotropic VH polarization data;
5-8, the detection performance of the invention for the small floating targets on the sea surface is superior to that of the existing fractal-based single-feature detection method.
Finally, it should be noted that: the above examples are only used to illustrate the technical solution of the present invention, but not to limit it; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.

Claims (3)

1. A sea surface floating small target detection method based on connected region characteristics is characterized in that: the method comprises the following steps:
step 1) constructing a U-Net network
The U-Net network is formed by repeatedly splicing a convolutional layer, a pooling layer and a regularization layer, the convolutional layer output is processed by using a ReLU activation function, and the size of a convolutional kernel is set to be 3 × 3 × nfAnd the number n of downsampling kernels at a timefDoubling the number of cores n sampled at a timefHalving;
step 2) making a U-Net network training data set
Step 2a) acquiring radar echo data:
sending a signal to the sea surface by using a radar transmitter, receiving echo data reflected by the sea surface by using a radar receiver, and acquiring radar echo data, wherein the echo data is divided into pure clutter data and echo data containing a target; selecting a part of distance units from the echo data containing the target as training units, wherein the time sequence of the training units is as follows: z (N), N ═ 1, 2.., N; taking the distance unit of the pure clutter as a reference unit, wherein the time sequence of the reference unit is as follows: z is a radical ofp(N), N1, 2, Q is the number of reference units, and N is the length of the time series;
step 2b) of comparing the training unit time series z with the reference unit time series zpCarrying out block whitening, and then iterating by using an approximate maximum likelihood estimation method to obtain a speckle covariance matrix of a clutter vector of each subinterval; carrying out block whitening on the training unit echo and the reference unit echo by using the iterated covariance matrix to obtain a whitened training unit echo vector and a whitened reference unit echo vector;
step 2c), making a whitened time-frequency diagram data set:
calculating a smooth Wigner-Willi distribution of the training units and a smooth Wigner-Willi distribution of the reference units; then storing the smooth Wigner-Willi distribution of the training unit and the smooth Wigner-Willi distribution of the reference unit as a time-frequency image format to obtain a whitened training unit time-frequency image data set and a whitened reference unit time-frequency image data set;
step 2d), manufacturing a training label corresponding to the whitening time-frequency diagram data set of the training unit:
marking a connected region where a target in a training unit time-frequency graph data set is located by using an image marking tool labelme, generating a json file after marking, and analyzing the json file to obtain a required label picture set;
step 3), training the U-Net network:
adopting the time-frequency graphs of 300 training units as a training set, adopting the time-frequency graphs of 30 training units as a verification set, and carrying out network training by using an Adam algorithm;
step 4), calculating detection statistics:
inputting the time-frequency graph into a U-Net network to obtain a connected region extraction graph in an RGB format, converting the connected region extraction graph into a binary image, setting a background gray value to be 0 and a target gray value to be 255, and counting the number of pixel points larger than 0 to obtain the area of the connected region, namely detecting a statistic D;
step 5), determining a detection threshold:
inputting a whitened time-frequency diagram data set of a reference unit containing pure clutter into a trained U-Net network, calculating a detection statistic D, and obtaining a detection threshold T through Monte Carlo simulation;
step 6), detection:
inputting the change time-frequency diagram of the unit to be detected into the trained U-Net network, and calculating a detection statistic D; wherein the unit to be detected is an unknown unit; comparing the detection statistic D with the detection threshold T, and judging the target detection according to the comparison result:
if D is larger than or equal to T, a target is considered;
if D < T, no target is considered.
2. The method for detecting the small sea surface floating target based on the connected region characteristics as claimed in claim 1, wherein the method comprises the following steps: the specific steps of the step 2b) are as follows:
training unit time sequence z and reference unit time sequence zpRespectively cut into short vectors with the length M not overlapping, namely:
z＝[z1,z2,...,zm,...,zN/M]T
zp＝[zp,1,zp,2,...,zp,m,...,zp,N/M]T,p＝1,2,...,Q
wherein the vector zmM short vectors, z, representing training unit time trainingp,mAn mth short vector representing a time sequence of the reference unit, wherein M is 1, 2.. and N/M, and a speckle covariance matrix of a clutter vector of each subinterval is obtained by iteration through an approximate maximum likelihood estimation method:
Figure FDA0002539947550000021
Figure FDA0002539947550000022
and (3) carrying out block whitening on the training unit echo and the reference unit echo by using the covariance matrix after iteration to obtain a whitened training unit echo vector and a whitened reference unit echo vector:
Figure FDA0002539947550000031
Figure FDA0002539947550000032
3. the method for detecting the small sea surface floating target based on the connected region characteristics as claimed in claim 2, wherein the method comprises the following steps: the specific steps of the step 2c) are as follows:
step 2c), making a whitened time-frequency diagram data set:
calculating a smooth Wigner-Willi distribution of the training units and a smooth Wigner-Willi distribution of the reference units:
Figure 1
Figure RE-FDA0002609681020000034
wherein the superscript denotes conjugation, g (m) is a time smoothing window, h (k) is a frequency smoothing window, E denotes half the length of the time smoothing window, F denotes half the length of the frequency smoothing window, Δ FdA sampling interval that is a normalized doppler frequency, N1, 2.. and N, l 1, 2.. and N;
and storing the time-frequency matrixes of the training units calculated by the smooth Wigner-Willi distribution and the smooth Wigner-Willi distribution of the reference units into an image format to obtain a whitened training unit time-frequency image data set.
CN202010543795.6A 2020-06-15 2020-06-15 Sea surface floating small target detection method based on connected region characteristics Active CN111707998B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010543795.6A CN111707998B (en) 2020-06-15 2020-06-15 Sea surface floating small target detection method based on connected region characteristics

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010543795.6A CN111707998B (en) 2020-06-15 2020-06-15 Sea surface floating small target detection method based on connected region characteristics

Publications (2)

Publication Number Publication Date
CN111707998A true CN111707998A (en) 2020-09-25
CN111707998B CN111707998B (en) 2023-03-31

Family

ID=72539958

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010543795.6A Active CN111707998B (en) 2020-06-15 2020-06-15 Sea surface floating small target detection method based on connected region characteristics

Country Status (1)

Country Link
CN (1) CN111707998B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112560844A (en) * 2020-12-09 2021-03-26 南京理工大学智能计算成像研究院有限公司 Automatic speckle pattern identification method
CN112784690A (en) * 2020-12-31 2021-05-11 西安电子科技大学 Broadband signal parameter estimation method based on deep learning
CN113009443A (en) * 2021-02-22 2021-06-22 南京邮电大学 Sea surface target detection method and device based on graph connection density

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2016097890A1 (en) * 2014-12-15 2016-06-23 Airbus Group Singapore Pte. Ltd. Automated method for selecting training areas of sea clutter and detecting ship targets in polarimetric synthetic aperture radar imagery
CN105738888A (en) * 2016-03-31 2016-07-06 西安电子科技大学 Double-characteristic sea surface floating small-target detection method based on sea clutter suppression
CN106569193A (en) * 2016-11-01 2017-04-19 西安电子科技大学 Sea-surface small target detection method based on front-back revenue reference particle filter

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2016097890A1 (en) * 2014-12-15 2016-06-23 Airbus Group Singapore Pte. Ltd. Automated method for selecting training areas of sea clutter and detecting ship targets in polarimetric synthetic aperture radar imagery
CN105738888A (en) * 2016-03-31 2016-07-06 西安电子科技大学 Double-characteristic sea surface floating small-target detection method based on sea clutter suppression
CN106569193A (en) * 2016-11-01 2017-04-19 西安电子科技大学 Sea-surface small target detection method based on front-back revenue reference particle filter

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
李东宸等: "块白化杂波抑制的海面漂浮小目标检测方法", 《西安电子科技大学学报》 *
许述文等: "基于四极化通道融合的海面漂浮微弱目标特征检测", 《信号处理》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112560844A (en) * 2020-12-09 2021-03-26 南京理工大学智能计算成像研究院有限公司 Automatic speckle pattern identification method
CN112784690A (en) * 2020-12-31 2021-05-11 西安电子科技大学 Broadband signal parameter estimation method based on deep learning
CN112784690B (en) * 2020-12-31 2022-12-27 西安电子科技大学 Broadband signal parameter estimation method based on deep learning
CN113009443A (en) * 2021-02-22 2021-06-22 南京邮电大学 Sea surface target detection method and device based on graph connection density
CN113009443B (en) * 2021-02-22 2023-09-12 南京邮电大学 Sea surface target detection method and device based on graph communication density

Also Published As

Publication number Publication date
CN111707998B (en) 2023-03-31

Similar Documents

Publication Publication Date Title
CN111707998B (en) Sea surface floating small target detection method based on connected region characteristics
Ao et al. Detection and discrimination of ship targets in complex background from spaceborne ALOS-2 SAR images
CN107229918B (en) SAR image target detection method based on full convolution neural network
Deng et al. Unsupervised segmentation of synthetic aperture radar sea ice imagery using a novel Markov random field model
CN109871902B (en) SAR small sample identification method based on super-resolution countermeasure generation cascade network
US6437728B1 (en) A-scan ISAR target recognition system and method
CN112766221B (en) Ship direction and position multitasking-based SAR image ship target detection method
Chen et al. Spatial–temporal convolutional gated recurrent unit network for significant wave height estimation from shipborne marine radar data
CN112541481A (en) Sea detection radar target detection method based on deep learning
Leung Applying chaos to radar detection in an ocean environment: an experimental study
CN110263646A (en) A kind of sea weak target detection method and system based on convolutional neural networks
CN113486961A (en) Radar RD image target detection method and system based on deep learning under low signal-to-noise ratio and computer equipment
CN105574529A (en) Target detection method of side scan sonar
Zong-ling et al. Remote sensing ship target detection and recognition system based on machine learning
CN115097396A (en) Radar active interference identification method based on CNN and LSTM series model
CN111707999B (en) Sea surface floating small target detection method based on combination of multiple features and ensemble learning
CN112835008B (en) High-resolution range profile target identification method based on attitude self-adaptive convolutional network
CN113887583A (en) Radar RD image target detection method based on deep learning under low signal-to-noise ratio
CN116740342A (en) Millimeter wave image target detection and identification method for improving YOLO v8
Ni et al. Multi-resolution fusion convolutional neural networks for intrapulse modulation LPI radar waveforms recognition
Arivazhagan et al. Optimal Gabor sub-band-based spectral kurtosis and Teager Kaiser energy for maritime target detection in SAR images
CN115223033A (en) Synthetic aperture sonar image target classification method and system
CN113807206B (en) SAR image target identification method based on denoising task assistance
CN115937977A (en) Few-sample human body action recognition method based on multi-dimensional feature fusion
CN104732190A (en) Synthetic aperture sonar target detection method based on orthogonal texture correlation analysis

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant