A ship classification and recognition method for low-resolution radar
By extracting eight-dimensional feature vectors and particle swarm optimization algorithms, the support vector machine classifier is optimized, which solves the problem of fewer categories and low accuracy in low-resolution radar ship target recognition, and achieves higher recognition Accuracy and robustness.
Patent Information
- Application Number
- CN202110642439.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-06-09
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2041-06-09
AI Technical Summary
Ship target recognition for low-resolution radars has the problem of fewer categories and insufficient accuracy.
By constructing the data set based on radar echo data and ship automatic identification system messages, eight-dimensional feature vectors are extracted, including contour image area, target lateral length, radar scattering cross-sectional area, echo amplitude peak, target velocity, image energy, image not Converting moment 1 and image invariant moment 2, and using the particle swarm optimization algorithm with linear decreasing inertia weights and learning factors, the support vector machine classifier with eight-dimensional features is trained for identification.
It improves the classification and identification accuracy of ship targets and increases the number of categories that can be divided, which has higher robustness and recognition capabilities than traditional methods.
Smart Images

Figure CN113378920B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of radar detection, and in particular relates to a ship classification and recognition method for low-resolution radar. Background Art
[0002] Maritime ship target recognition is of great significance in military fields such as maritime monitoring and military reconnaissance, as well as in civilian fields such as maritime search and rescue and maintaining sea traffic. Since high-resolution radars have a long imaging time, high research costs and cycle requirements, and most of the radars in service are narrow-band low-resolution radars, ship target recognition for low-resolution radars has a high application value.
[0003] Low-resolution radar has a narrow bandwidth, poor radial distance resolution, and radar echo sequence features are prone to aliasing. Compared with the one-dimensional echo sequence, the two-dimensional profile image composed of distance and azimuth not only provides echo amplitude information, but also reflects its correlation information with the azimuth. Therefore, extracting radar target features based on the two-dimensional profile image of the echo sequence provides favorable auxiliary information for more accurate target identification. At the same time, the extraction of multi-dimensional features can reflect the various characteristics of the target from different angles, thereby improving the accuracy of target classification.
[0004] Currently, the ship target recognition for low-resolution radar has the problems of too few classification categories and low accuracy. Summary of the invention
[0005] The present invention aims to provide a ship classification and recognition method for low-resolution radar, so as to solve the technical problems of a small number of classification categories and insufficient accuracy.
[0006] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows:
[0007] A ship classification and recognition method for low-resolution radar includes the following steps:
[0008] Step 1: construct a data set based on radar echo data and ship automatic identification system messages;
[0009] Step 2, extracting eight-dimensional feature vectors; wherein the eight-dimensional features include contour image area, target lateral length, radar cross-sectional area, echo amplitude peak, target speed, image energy, image invariant moment 1 and image invariant moment 2; the eight-dimensional feature vector extraction reflects the characteristics of the ship target from five angles: target shape, radar cross-sectional area, echo amplitude, target speed attribute and two-dimensional grayscale image attribute, so as to improve the classification and recognition accuracy;
[0010] Step 3: Optimization and training methods of classifier parameters;
[0011] Step 4: Classify and identify radar targets.
[0012] Furthermore, the classifier training and parameter optimization method in step 3 includes the following steps:
[0013] Step 3.1, normalize the feature vector to obtain a feature vector set for training;
[0014] Step 3.2, based on the feature vector set, the PSO algorithm with linear decreasing inertia weight and learning factor is used to optimize the radial basis kernel function parameter γ and penalty factor C of the classifier;
[0015] Step 3.3, training a classifier based on the feature vector set;
[0016] Step 3.4: Save the normalized parameters and the trained model.
[0017] Furthermore, the classification and identification of radar targets in step 4 includes the following steps:
[0018] Step 4.1, normalize the feature vector using the normalization parameters saved in step 3;
[0019] Step 4.2: Use the model trained in step 3 to perform classification and recognition;
[0020] Step 4.3: Output classification results and confidence levels. The ship classification and recognition method for low-resolution radar of the present invention has the following advantages:
[0021] (1) The present invention constructs a ship target data set based on the established C-band radar echo data and AIS messages, which facilitates subsequent feature extraction and training work.
[0022] (2) The present invention extracts eight features based on the echo sequence contour image to form a feature vector, which can reflect various characteristics of ship targets from five angles: target shape, radar scattering cross-sectional area, echo amplitude, target speed attribute and two-dimensional grayscale image attribute. Compared with the feature combination used in existing research, the number of separable categories and recognition accuracy are further improved.
[0023] (3) The present invention uses an SVM classifier based on eight-dimensional features. Compared with the nearest neighbor (K-Nearest Neighbor, KNN) algorithm, neural network algorithm, etc., it has the characteristics of strong robustness, low complexity, and good generalization ability, and is more suitable for low-resolution radar target recognition. At the same time, the particle swarm optimization (PSO) algorithm is used to optimize the parameters of the classifier, which has a faster optimization speed and improves the classification ability. The recognition accuracy is higher than the traditional grid search method. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 is a data processing flow chart of the present invention;
[0025] Figure 2 It is a schematic diagram of the data format collected by the present invention;
[0026] FIG3( a ) is a two-dimensional echo sequence profile image of a ship of the present invention having a length of 97 m and a distance of 8 km from the radar;
[0027] FIG3( b ) is a two-dimensional echo sequence profile image of a ship of the present invention having a length of 165 m and a distance of 8 km from the radar;
[0028] FIG3( c ) is a two-dimensional echo sequence profile image of a ship of the present invention having a length of 183 m and a distance of 8 km from the radar;
[0029] FIG3( d ) is a two-dimensional echo sequence profile image of a ship of the present invention having a length of 97 m and a distance of 4 km from the radar;
[0030] FIG3( e ) is a two-dimensional echo sequence profile image of a ship of the present invention having a length of 115 m and a distance of 4 km from the radar;
[0031] FIG3( f ) is a two-dimensional echo sequence profile image of a ship of the present invention having a length of 143 m and a distance of 4 km from the radar; DETAILED DESCRIPTION
[0032] In order to better understand the purpose, structure and function of the present invention, the following is a further detailed description of a ship classification and identification method for low-resolution radar of the present invention in conjunction with the accompanying drawings.
[0033] The implementation process of the present invention includes two parts: classifier training and ship target recognition to be tested:
[0034] For classifier training, first collect radar echo sequence profile data, construct a data set based on echo data and ship automatic identification system (Automatic Identification System, AIS) message; extract eight features to form a feature vector set; normalize the feature vector and save the normalization parameters; based on the normalized feature vector set, optimize and train the support vector machine (SVM) classifier parameters, and save the trained model.
[0035] For the target identification of the ship to be tested, the target data of the ship to be tested is first collected; eight features are extracted to form a feature vector; the feature vector is normalized using normalization parameters; the ship is classified and judged by length using an SVM classifier based on eight-dimensional features; and the classification results and confidence levels are output.
[0036] The eight-dimensional features include: contour image area, target lateral length, radar cross-sectional area, echo amplitude peak, target speed, image energy, image invariant moment 1 and image invariant moment 2. Normalization parameters include mean and variance.
[0037] like Figure 1 Shown is a data processing flow chart of the present invention.
[0038] The present invention comprises the following steps:
[0039] Step 1: Construct the data set:
[0040] The data collection process is as follows: use the C-band radar that has been set up, start the data collector to record the radar echo data; at the same time, obtain the target position, motion status, target type and other information based on the echo data and the ship automatic identification system (Automatic Identification System, AIS) message; the original data is stored in DAT files, each file includes two parts: the maritime mobile service identification code (Maritime Mobile Service Identify, MMSI), start / end azimuth, start / end distance and other ship information and the echo sequence. Figure 2 As shown in Figure 3, the file format is as follows: the first 42 bytes are the header file including the ship information, and the echo sequence is arranged in the form of "azimuth + echo value" in order of azimuth. As shown in Figure 3(a), Figure 3(b), Figure 3(c), Figure 3(d), Figure 3(e), and Figure 3(f), the ship length information can be searched according to the unique MMSI number of each ship. The original data is classified by MMSI number as the training and test data sets.
[0041] Step 2: Eight-dimensional feature extraction.
[0042] Step 2.1: Extract shape features: The number of non-zero points of the echo sequence contour image (i.e., the target contour image area) is used as the first feature f1, and the target transverse length is used as the second feature f2.
[0043]
[0044] f2=θ T R
[0045] Among them, y i Indicates the magnitude of the i-th echo value of the radar target, θ T is the width caused by the lateral length of the target, and R is the distance between the target and the radar.
[0046] Step 2.2: Extract the radar cross-section as the third feature f3, expressed as follows:
[0047] f3=10log10 (P r R 4 )
[0048] Among them, P r is the target echo power received by the radar.
[0049] Step 2.3, extract the echo amplitude peak as the fourth feature f4, expressed as follows:
[0050] f4=max{y i}
[0051] Step 2.4: Extract the target speed as the fifth feature f5. The currently used low-resolution radar can generally directly measure the target speed. The target speed is represented by v:
[0052] f5=v
[0053] Step 2.5, extract the features of the two-dimensional grayscale image. First, normalize the image and convert the target contour image into a general two-dimensional grayscale image. Let H(i,j) represent the pixel value of the i-th row and j-th column of the grayscale image. Then the sixth feature is the image energy f6, which is expressed as:
[0054]
[0055] Let (x, y) be the coordinates of each point on the object represented by the binary image, (x c ,y c ) is the center of the object, then the p+qth order central moment is expressed as
[0056]
[0057] where a1…a n is a constant.
[0058] The seventh and eighth features used are invariant moment 1 and invariant moment 2, respectively, expressed as
[0059] f7=μ 2,0 +μ 0,2
[0060]
[0061] Step 3: Classifier training and parameter optimization method:
[0062] Step 3.1, construct the training set according to step 1;
[0063] Step 3.2, the eight features extracted in step 2 form a feature vector;
[0064] Step 3.3, normalize the feature vector to obtain a feature vector set for training;
[0065] Step 3.4, based on the feature vector set, the PSO algorithm with linear decreasing inertia weight and learning factor is used to optimize the radial basis kernel function (RBF) parameter γ and penalty factor C of the SVM classifier;
[0066] Step 3.5: training a classifier based on the feature vector set;
[0067] Step 3.6. Save the normalized parameters and the trained model.
[0068] Step 4: Radar target classification and identification:
[0069] Step 4.1, the eight features extracted from the test data collected in step 1 according to step 2 form a feature vector;
[0070] Step 4.2, normalize the feature vector using the normalization parameters saved in step 3;
[0071] Step 4.3: Use the model trained in step 3 to perform classification and recognition;
[0072] Step 4.4: Output the classification results and confidence levels.
[0073] It is to be understood that the present invention is described by some embodiments, and it is known to those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the scope of protection of the present invention.
Claims
1. A ship classification and recognition method for low-resolution radar, characterized in that: The following steps are involved: Step 1: construct a data set based on radar echo data and ship automatic identification system messages; Step 2, extracting eight-dimensional feature vectors; wherein the eight-dimensional features include contour image area, target lateral length, radar cross-sectional area, echo amplitude peak, target speed, image energy, image invariant moment 1 and image invariant moment 2; the eight-dimensional feature vector extraction reflects the characteristics of the ship target from five angles: target shape, radar cross-sectional area, echo amplitude, target speed attribute and two-dimensional grayscale image attribute, so as to improve the classification and recognition accuracy; Step 2: Eight-dimensional feature extraction; Step 2.1, extract shape features; the number of non-zero points of the echo sequence contour image is used as the first feature f1, and the horizontal length of the target is used as the second feature f2; f2=θ T R Among them, y i Indicates the magnitude of the i-th echo value of the radar target, θ T is the width caused by the lateral length of the target, and R represents the distance between the target and the radar; Step 2.2: Extract the radar cross-section as the third feature f3, expressed as follows: f3=10log 10 (P r R 4 ) Among them, P r is the target echo power received by the radar; Step 2.3, extract the echo amplitude peak as the fourth feature f4, expressed as follows: f4=max{y i } Step 2.4, extract the target speed as the fifth feature f5; the target speed is represented by v: f5=v Step 2.5, extract the two-dimensional grayscale image features; first normalize the image and convert the target contour image into a two-dimensional grayscale image; use H(i,j) to represent the pixel value of the i-th row and j-th column of the grayscale image, then the sixth feature is the image energy f6, which is expressed as: Let (x, y) be the coordinates of each point on the object represented by the binary image, (x c ,y c ) is the center of the object, then the p+qth order central moment is expressed as where a1…a n is a constant; The seventh and eighth features used are invariant moment 1 and invariant moment 2, respectively, expressed as f7=μ 2,0 +m 0,2 Step 3: Optimization and training methods of classifier parameters; Step 4: Classify and identify radar targets; The classifier training and parameter optimization method in step 3 includes the following steps: Step 3.1, normalize the feature vector to obtain a feature vector set for training; Step 3.2, based on the feature vector set, the PSO algorithm with linear decreasing inertia weight and learning factor is used to optimize the radial basis kernel function parameter γ and penalty factor C of the classifier; Step 3.3, training a classifier based on the feature vector set; Step 3.4: Save the normalized parameters and the trained model.
2. The ship classification and identification method for low-resolution radar according to claim 1 is characterized in that: Classifying and identifying radar targets in step 4 includes the following steps: Step 4.1, normalize the feature vector using the normalization parameters saved in step 3; Step 4.2: Use the model trained in step 3 to perform classification and recognition; Step 4.3: Output the classification results and confidence levels.
