CN111539385A - Extremely narrow pulse radar ship identification method based on resolution pyramid model - Google Patents
Extremely narrow pulse radar ship identification method based on resolution pyramid model Download PDFInfo
- Publication number
- CN111539385A CN111539385A CN202010466321.6A CN202010466321A CN111539385A CN 111539385 A CN111539385 A CN 111539385A CN 202010466321 A CN202010466321 A CN 202010466321A CN 111539385 A CN111539385 A CN 111539385A
- Authority
- CN
- China
- Prior art keywords
- slice
- target
- radar
- classification
- classifier
- 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
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
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/88—Radar or analogous systems specially adapted for specific applications
- G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
- G01S13/90—Radar or analogous systems specially adapted for specific applications for mapping or imaging using synthetic aperture techniques, e.g. synthetic aperture radar [SAR] techniques
- G01S13/9021—SAR image post-processing techniques
- G01S13/9027—Pattern recognition for feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- 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
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/88—Radar or analogous systems specially adapted for specific applications
- G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
-
- 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
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/88—Radar or analogous systems specially adapted for specific applications
- G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
- G01S13/90—Radar or analogous systems specially adapted for specific applications for mapping or imaging using synthetic aperture techniques, e.g. synthetic aperture radar [SAR] techniques
- G01S13/9094—Theoretical aspects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10044—Radar image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Engineering & Computer Science (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Electromagnetism (AREA)
- Computer Networks & Wireless Communication (AREA)
- Multimedia (AREA)
- Probability & Statistics with Applications (AREA)
- Astronomy & Astrophysics (AREA)
- Image Analysis (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a method for identifying a ship by using a narrow pulse radar based on a resolution pyramid model, which can be used for identifying ships stably, reliably and more accurately by using radar images acquired by the narrow pulse radar under the condition of a small sample. Segmenting a ship target in the radar image from the background to obtain a ship target image; extracting distinguishing slice characteristic vectors by an unsupervised method, clustering all distinguishing slices, constructing a characteristic dictionary to encode all the slice characteristic vectors of a single radar image, quantizing each slice characteristic vector into a representation vector of the distinguishing slice closest to the Euclidean distance of the slice characteristic vector, summing all the slice characteristic vectors, and normalizing the sum to obtain a middle-layer semantic characteristic vector of the radar image; and processing the radar images at the historical moment to obtain a middle-layer semantic feature vector serving as a target classification training set, training an SVM classifier to obtain a target classifier, and identifying the ship target by adopting the target classifier.
Description
Technical Field
The invention relates to the technical field of ship target detection, in particular to a very narrow pulse radar ship identification method based on a resolution pyramid model.
Background
In recent years, ship detection and identification by using radar images are highly regarded in the field of marine remote sensing application. The radar can observe a large-range ocean area all day long and all weather, and is one of effective means for identifying ships in the large-range ocean area.
At present, ship identification mainly comprises the following two modes:
firstly, the method is used for identifying based on various primary image low-layer characteristics such as geometric structures and electromagnetic scattering, and the method is simple, but the characteristics are variable and the identification effect is poor. When the attitude, the geometric form and the radar parameter of the ship change and the change of the low-level features is large, the description of the ship on the target is easily influenced by the external environment and has certain instability when the bottom-level features are adopted for target identification.
And secondly, ship target identification based on deep learning, the identification algorithm does not need artificial image preprocessing, feature extraction and classifier design, but the method needs a large number of training samples. For the ultra-narrow pulse radar, the duration of the ultra-narrow pulse is short, generally in ns level, so that the signal bandwidth is large and reaches GHz, and therefore, the number of ship images acquired by the ultra-narrow pulse radar is small, and the ultra-narrow pulse radar is not suitable for the mode.
Therefore, how to perform stable and reliable ship identification with higher accuracy on radar images acquired by the extremely narrow pulse radar under the condition of small samples is a problem to be solved urgently at present.
Disclosure of Invention
In view of this, the invention provides a method for identifying a ship by using a very narrow pulse radar based on a resolution pyramid model, which can identify ships stably, reliably and with higher accuracy by using radar images acquired by the very narrow pulse radar under the condition of a small sample.
In order to achieve the purpose, the technical scheme of the invention is as follows: the extremely narrow pulse radar ship identification method based on the resolution pyramid model comprises the following steps:
And 2, extracting the characteristic vector of the discriminant slice by adopting an unsupervised method aiming at the ship target image.
The method specifically comprises the following steps:
step 201, carrying out multi-scale slicing on the ship target image.
The multi-scale slicing procedure is as follows: presetting a plurality of slice scales, carrying out multi-scale slicing on a ship target image to obtain a certain number of slice images, and extracting the HOG (histogram of oriented gradients) characteristic as a characteristic vector of each slice image.
Carrying out dimension unified processing on the feature vectors of all the slice images of all the ship target images to ensure that the feature vectors of all the slice images have the same dimension, taking the feature vectors of the slice images as slice samples, and randomly and uniformly dividing all the slice samples into a slice classification training set and a slice classification verification set; setting a maximum iteration number N; and setting the initial value of the current iteration count i as 1.
Step 202, clustering the slice classification training set to obtain a cluster group of the slice classification training set.
Step 203, training a corresponding SVM classifier, namely a class group classifier, aiming at each clustering class group, wherein a positive sample of the class group classifier is a class group member of the current clustering class group, and a negative sample is a training set sample without the positive sample.
Step 204, classifying the slice classification verification set by using the class group classifier trained in the step 203 to obtain a classifier classification result; and simultaneously carrying out k-means clustering processing on the slice classification verification set to obtain a clustering result.
And calculating the accuracy of the classification result of the classifier by taking the clustering result as a standard, and removing the class members of the class classifier corresponding to the classification result of the classifier with the lowest accuracy.
Step 205, exchanging the slice classification validation set and the slice classification training set, and judging whether the maximum iteration number N is reached.
If yes, all slice samples remaining after the removal operation in step 204 are output as discriminant slice feature vectors, and step 3 is performed.
If not, the process returns to step 202 to continue execution.
And 3, clustering all the discriminant slice feature vectors, and constructing a clustering center as a feature dictionary.
And 4, acquiring the middle-layer semantic feature vector of each radar image by using the feature dictionary for all the acquired radar images.
The specific process is as follows: and coding all slice feature vectors of the same radar image by using a feature dictionary, quantizing each slice feature vector into a representation vector of a discriminant slice closest to the Euclidean distance of the slice feature vector, and summing and normalizing all the slice feature vectors to obtain a middle-layer semantic feature vector of the current radar image.
And 5, taking the middle-layer semantic feature vectors of all the acquired radar images as a target classification training set, carrying out target class label setting on all the middle-layer semantic feature vectors in the target classification training set, and training an SVM classifier by using the target classification training set to obtain a target classifier.
And 6, taking the newly acquired radar image to be classified, acquiring a middle-layer semantic feature vector of the radar image to be classified by using the feature dictionary, and inputting the middle-layer semantic feature vector to a target classifier to perform ship target identification.
Further, the ship target in the radar image is segmented from the background to obtain a ship target image, which specifically comprises: removing speckle noise in the radar image by using a denoising algorithm, determining a main shaft and a segmentation range of the ship target by using Radon transformation, and finally segmenting the ship target from the background of the radar image to obtain a ship target image.
Further, clustering is performed on all the discriminant slices, and a clustering center is constructed as a feature dictionary, specifically: clustering all the discriminant slice feature vectors to obtain N cluster centers, wherein the nth cluster center vector is dnN is 1,2, …, N; the feature dictionary is then: d ═ D1,d2,...,dN]。
Further, for all the acquired radar images, a feature dictionary is used to acquire a middle-layer semantic feature vector of each radar image, specifically:
for the same radar image, with fiFor the feature vector of the ith slice image, the feature vectors of all slice images are encoded by using a feature dictionary to obtain a corresponding expression vector V based on a middle-layer semantic feature dictionaryi=[v1,v2,…,vN]Represents the nth element v in the vectornThe encoding rule of (1) is as follows:
||fi-dk||2denotes fiAnd dkThe Euclidean distance of (c); k denotes 1,2,3 …, the numerical value in N.
And summing and normalizing the expression vectors of the feature vectors of all the slice images based on the middle-layer semantic feature dictionary to obtain the middle-layer semantic feature vector of the radar image.
Further, step 4 further comprises: processing radar images at part of historical moments in steps 1-3 to obtain middle-layer semantic feature vectors serving as a target classification test set; and (3) performing target class label setting on all middle-layer semantic feature vectors in the target classification test set, performing classification test on the target classification test set by using a trained target classifier, calculating the classification accuracy of the target classifier, retraining the feature dictionary if the classification accuracy of the target classifier is lower than a set threshold, and executing the steps 5 and 6.
Has the advantages that:
the invention provides a resolution pyramid model-based extremely narrow pulse radar ship identification method, which comprises the steps of firstly carrying out multi-scale slicing on a segmented ship target image, adopting an unsupervised method to extract discriminative slice characteristic vectors, wherein the corresponding slices are discriminative visual semantic slices, and providing a processing basis for later middle-layer semantic feature extraction; and then, analyzing and processing the low-level features of the radar image, namely the HOG features, to construct a middle-level semantic feature vector, wherein the middle-level semantic feature vector can describe the target more stably relative to the low-level features and is not influenced by the outside as much as possible, and the extracted middle-level semantic feature vector can realize stable and reliable ship identification by adopting an SVM (support vector machine) method.
Drawings
FIG. 1 is a flow chart of a method for identifying a ship by using a narrow pulse radar based on a resolution pyramid model, provided by the invention;
FIG. 2 is a schematic view of unsupervised extraction of a target discriminative slice;
FIG. 3 is a schematic diagram of a two-dimensional visualization of sample HOG features;
FIG. 4 is a diagram of a middle level semantic feature representation of a picture of a ship;
FIG. 5 is a diagram of a sample two-dimensional visualization based on a middle level semantic dictionary representation.
Detailed Description
The invention is described in detail below by way of example with reference to the accompanying drawings.
The invention provides a method for identifying a ship by using a narrow pulse radar based on a resolution pyramid model, which comprises the following steps as shown in figure 1:
step 201, carrying out multi-scale slicing on a ship target image;
the multi-scale slicing procedure is as follows: presetting a plurality of slice scales, and carrying out multi-scale slicing on the ship target image to obtain a certain number of slice images, wherein the slice scales can be set according to the sizes of pictures, for example, the slice scales are 8 multiplied by 8, 12 multiplied by 12, and the step is 4; and extracting the HOG (histogram of oriented gradients) characteristic as a characteristic vector of the slice image for each slice image.
Carrying out dimension unified processing on the feature vectors of all the slice images, carrying out dimension unified processing on the feature vectors by using a dimension reduction algorithm (such as a PCA dimension reduction algorithm and a TSNE algorithm), enabling the feature vectors of all the slice images to have the same dimension, taking the feature vectors of the slice images as slice samples, and randomly and uniformly dividing all the slice samples into a slice classification training set and a slice classification verification set; setting a maximum iteration number N; and setting the initial value of the current iteration count i as 1.
Step 202, clustering the slice classification training set to obtain a cluster group of the slice classification training set. The number of cluster groups is determined according to the number of samples of the training set, generally being one percent of the number of training samples, and the cluster samples with the number of cluster members being less than 0.75 times of the average number of cluster members are removed.
Step 203, training a corresponding SVM classifier, namely a cluster classifier, for each cluster group, wherein a positive sample of the cluster classifier is a cluster member of the current cluster group, and a negative sample is a training set sample without the positive sample, so that in order to avoid the overlarge difference in the number of the positive and negative samples, the negative sample can be properly reduced, and one tenth of the training set without the positive sample is generally selected as the negative sample.
Step 204, classifying the slice classification verification set by using the class group classifier trained in the step 203 to obtain a classifier classification result; and simultaneously carrying out k-means clustering processing on the slice classification verification set to obtain a clustering result.
And calculating the accuracy of the classification result of the classifier by taking the clustering result as a standard, and removing the class members of the class classifier corresponding to the classification result of the classifier with the lowest accuracy.
Step 205, exchanging the slice classification validation set and the slice classification training set, and judging whether the maximum iteration number N is reached:
if yes, all slice samples remaining after the removal operation in step 204 are output as discriminant slice feature vectors, and step 3 is performed.
If not, the process returns to step 202 to continue execution.
The iterative mode of the steps 2020-205 can effectively remove some cluster samples with the worst classification accuracy, so that the final output classification accuracy is higher.
And 3, clustering all the discriminant slices, and constructing a clustering center as a feature dictionary.
The step 3 specifically comprises the following steps: clustering all the discriminant slices to obtain N cluster centers, wherein the vector of the nth cluster center is dn,n=1,2,…,N。
The feature dictionary is then: d ═ D1,d2,...,dN]。
And 4, acquiring the middle-layer semantic feature vector of each radar image by using the feature dictionary for all the acquired radar images.
The specific process is as follows: and coding all slice feature vectors of the same radar image by using a feature dictionary, quantizing each slice feature vector into a representation vector of a discriminant slice closest to the Euclidean distance of the slice feature vector, and summing and normalizing all the slice feature vectors to obtain a middle-layer semantic feature vector of the current radar image.
In the embodiment of the invention, f is used for the same radar imageiFor the feature vector of the ith slice image of the image, the feature vectors of all slice images are encoded by using a feature dictionary to obtain a corresponding expression vector V based on a middle-layer semantic feature dictionaryi=[v1,v2,…,vN]Represents the nth element v in the vectornThe encoding rule of (1) is as follows:
||fi-dk||2denotes fiAnd dkThe Euclidean distance of (c); k denotes 1,2,3 …, the numerical value in N;
and summing and normalizing the expression vectors of the feature vectors of all the slice images based on the middle-layer semantic feature dictionary to obtain the middle-layer semantic feature vector of the current radar image.
Step 5, taking the middle-layer semantic feature vectors of all the acquired radar images as a target classification training set, performing target class label setting on all the middle-layer semantic feature vectors in the target classification training set, and training an SVM classifier by using the target classification training set to obtain a target classifier;
and 6, taking the newly acquired radar image to be classified, acquiring a middle-layer semantic feature vector of the radar image to be classified by using the feature dictionary, and inputting the middle-layer semantic feature vector to a target classifier to perform ship target identification.
The test set tests the trained SVM classifier, calculates the accuracy,
The effects of the present invention are further illustrated by the following experiments with measured data:
experimental scenarios and parameters
The data used in the experiment were TerraSAR datasets, with 15 each for bulk carriers, container ships, tankers, with 10 each for each type of target as training samples and 5 each as test samples. The results of two-dimensional visualization of the extracted HOG features after sample preprocessing are shown in fig. 3(a) and (b).
Contents and results of the experiments
The method provided by the invention is used for extracting the discriminant slices from the three types of data images, and the slices extracted by the unsupervised method have certain semantic and discriminant characteristics, such as the hull part of a bulk cargo ship, the container part of a container ship and the oil pipeline part of an oil tanker, and the three types of ship targets can be well distinguished by using the parts, so that the ship targets can be stably expressed by constructing a middle-layer semantic dictionary to encode pictures by using the low-layer HOG characteristics corresponding to the parts.
The middle-layer semantic feature expression results of the training samples and the test samples can be obtained through the image expression processing flow based on the middle-layer semantic dictionary, and the middle-layer semantic feature vector of a certain sample is selected and shown in fig. 4.
Fig. 5(a) and (b) are two-dimensional visualization results of extracted middle-layer semantic features, it can be found that there are clear class boundary lines between three types of ship targets represented by the middle-layer semantic feature dictionary, and the feature distributions of the training sample and the test sample are substantially the same. In order to ensure the reliability of the test result, a method of averaging the test results for multiple times is adopted as the final discrimination rate. Experiments show that the target identification rate based on the middle-layer semantic features is 86.67%. Through the middle-layer semantic feature visualization result and experimental data description, clear class boundary lines exist among the three types of ships represented based on the middle-layer semantic feature, the feature distribution is stable, and the performance of the classifier trained by utilizing the middle-layer semantic feature is obviously improved.
In summary, the above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (5)
1. The extremely narrow pulse radar ship identification method based on the resolution pyramid model is characterized by comprising the following steps of:
step 1, taking more than 2 radar images at historical time, and segmenting a ship target in the radar images from a background to obtain a ship target image;
step 2, extracting a discriminant slice feature vector by adopting an unsupervised method aiming at a ship target image;
the method specifically comprises the following steps:
step 201, carrying out multi-scale slicing on a ship target image;
the multi-scale slicing process is as follows: presetting a plurality of slice scales, carrying out multi-scale slicing on the ship target image to obtain a certain number of slice images, and extracting directional gradient Histogram (HOG) features as feature vectors of the slice images aiming at each slice image;
carrying out dimension unified processing on the feature vectors of all the slice images of all the ship target images to ensure that the feature vectors of all the slice images have the same dimension, taking the feature vectors of the slice images as slice samples, and randomly and uniformly dividing all the slice samples into a slice classification training set and a slice classification verification set; setting a maximum iteration number N; setting an initial value of a current iteration count i as 1;
202, clustering the slice classification training set to obtain a clustering group of the slice classification training set;
step 203, training a corresponding SVM classifier, namely a class group classifier, aiming at each clustering class group, wherein positive samples of the class group classifier are class group members of the current clustering class group, and negative samples are training set samples except the positive samples;
step 204, classifying the slice classification verification set by using the class group classifier trained in the step 203 to obtain a classifier classification result; meanwhile, carrying out k-means clustering processing on the slice classification verification set to obtain a clustering result;
calculating the accuracy of the classification result of the classifier by taking the clustering result as a standard, and removing the class members of the class classifier corresponding to the classification result of the classifier with the lowest accuracy;
step 205, exchanging the slice classification validation set and the slice classification training set, and judging whether the maximum iteration number N is reached:
if yes, outputting all slice samples left after the removing operation in the step 204 as discriminant slice feature vectors, and performing a step 3;
if not, returning to the step 202 to continue execution;
step 3, clustering all the discriminant slice feature vectors, and constructing a clustering center as a feature dictionary;
step 4, for all the obtained radar images, utilizing the feature dictionary to obtain a middle-layer semantic feature vector of each radar image;
the specific process is as follows: coding all slice feature vectors of the same radar image by using the feature dictionary, quantizing each slice feature vector into a representation vector of a discriminant slice closest to the Euclidean distance of the slice feature vector, and summing and normalizing all the slice feature vectors to be used as a middle-layer semantic feature vector of the current radar image;
step 5, taking the middle-layer semantic feature vectors of all the acquired radar images as a target classification training set, performing target class label setting on all the middle-layer semantic feature vectors in the target classification training set, and training an SVM classifier by using the target classification training set to obtain a target classifier;
and 6, taking the newly acquired radar image to be classified, acquiring a middle-layer semantic feature vector of the radar image to be classified by using the feature dictionary, and inputting the middle-layer semantic feature vector to the target classifier to perform ship target identification.
2. The method according to claim 1, wherein the segmentation of the ship target in the radar image from the background to obtain the ship target image comprises:
removing speckle noise in the radar image by using a denoising algorithm, determining a main shaft and a segmentation range of the ship target by using Radon transformation, and finally segmenting the ship target from the background of the radar image to obtain a ship target image.
3. The method according to claim 1, wherein the clustering is performed on all discriminant slices and the cluster centers are constructed as a feature dictionary, specifically:
clustering all the discriminant slice feature vectors to obtain N cluster centers, wherein the nth cluster center vector is dn,n=1,2,…,N;
The feature dictionary is then: d ═ D1,d2,...,dN]。
4. The method according to claim 3, wherein the step of obtaining the middle-level semantic feature vector of each radar image by using the feature dictionary for all the obtained radar images comprises:
for the same radar image, with fiFor the feature vector of the ith slice image, the feature vectors of all slice images are encoded by using a feature dictionary to obtain a corresponding expression vector V based on a middle-layer semantic feature dictionaryi=[v1,v2,…,vN]Represents the nth element v in the vectornThe encoding rule of (1) is as follows:
||fi-dk||2denotes fiAnd dkThe Euclidean distance of (c); k denotes 1,2,3 …, the numerical value in N;
and summing and normalizing the expression vectors of the feature vectors of all the slice images based on the middle-layer semantic feature dictionary to obtain the middle-layer semantic feature vector of the current radar image.
5. The method of claim 1, wherein step 4 further comprises: taking a middle-layer semantic feature vector of a part of acquired radar images as a target classification test set; and (3) performing target class label setting on all middle-layer semantic feature vectors in the target classification test set, performing classification test on the target classification test set by using a trained target classifier, calculating the classification accuracy of the target classifier, retraining the feature dictionary if the classification accuracy of the target classifier is lower than a set threshold, and executing the steps 5 and 6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010466321.6A CN111539385B (en) | 2020-05-28 | 2020-05-28 | Extremely narrow pulse radar ship identification method based on resolution pyramid model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010466321.6A CN111539385B (en) | 2020-05-28 | 2020-05-28 | Extremely narrow pulse radar ship identification method based on resolution pyramid model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111539385A true CN111539385A (en) | 2020-08-14 |
CN111539385B CN111539385B (en) | 2022-05-20 |
Family
ID=71978011
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010466321.6A Active CN111539385B (en) | 2020-05-28 | 2020-05-28 | Extremely narrow pulse radar ship identification method based on resolution pyramid model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111539385B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112198488A (en) * | 2020-09-08 | 2021-01-08 | 北京理工大学 | Extremely narrow pulse radar angle-associated target echo space aggregation method |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2003090879A (en) * | 2001-09-18 | 2003-03-28 | Toshiba Corp | Category identification method of radar targets, computer program and device therefor |
CN106096506A (en) * | 2016-05-28 | 2016-11-09 | 重庆大学 | Based on the SAR target identification method differentiating doubledictionary between subclass class |
CN106408030A (en) * | 2016-09-28 | 2017-02-15 | 武汉大学 | SAR image classification method based on middle lamella semantic attribute and convolution neural network |
CN108256471A (en) * | 2018-01-16 | 2018-07-06 | 北方工业大学 | High-resolution synthetic aperture radar ship identification method based on double-layer classification network |
-
2020
- 2020-05-28 CN CN202010466321.6A patent/CN111539385B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2003090879A (en) * | 2001-09-18 | 2003-03-28 | Toshiba Corp | Category identification method of radar targets, computer program and device therefor |
CN106096506A (en) * | 2016-05-28 | 2016-11-09 | 重庆大学 | Based on the SAR target identification method differentiating doubledictionary between subclass class |
CN106408030A (en) * | 2016-09-28 | 2017-02-15 | 武汉大学 | SAR image classification method based on middle lamella semantic attribute and convolution neural network |
CN108256471A (en) * | 2018-01-16 | 2018-07-06 | 北方工业大学 | High-resolution synthetic aperture radar ship identification method based on double-layer classification network |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112198488A (en) * | 2020-09-08 | 2021-01-08 | 北京理工大学 | Extremely narrow pulse radar angle-associated target echo space aggregation method |
Also Published As
Publication number | Publication date |
---|---|
CN111539385B (en) | 2022-05-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108122008B (en) | SAR image recognition method based on sparse representation and multi-feature decision-level fusion | |
US9558550B2 (en) | Method and system for the automatic analysis of an image of a biological sample | |
US10853937B2 (en) | Unsupervised image-based anomaly detection using multi-scale context-dependent deep autoencoding gaussian mixture model | |
Awad | An Unsupervised Artificial Neural Network Method for Satellite Image Segmentation. | |
US9196044B2 (en) | False alarm rejection for boat detection candidates | |
Chehrsimin et al. | Automatic individual identification of Saimaa ringed seals | |
Beyan et al. | Detection of abnormal fish trajectories using a clustering based hierarchical classifier | |
Liu et al. | An improved InceptionV3 network for obscured ship classification in remote sensing images | |
CN109657704B (en) | Sparse fusion-based coring scene feature extraction method | |
Elnemr | Feature selection for texture-based plant leaves classification | |
CN111539385B (en) | Extremely narrow pulse radar ship identification method based on resolution pyramid model | |
WO2017124221A1 (en) | System and method for object detection | |
Ramanathan et al. | A novel technique for english font recognition using support vector machines | |
Chen et al. | Ship target discrimination in SAR images based on BOW model with multiple features and spatial pyramid matching | |
CN111767806B (en) | Ultra-narrow pulse radar ship target identification method based on Attribute | |
CN111832463A (en) | Deep learning-based traffic sign detection method | |
US11461891B2 (en) | Phenotyping tumor infiltrating lymphocytes on hematoxylin and eosin (HandE) stained tissue images to predict recurrence in lung cancer | |
CN109558771B (en) | Behavior state identification method, device and equipment of marine ship and storage medium | |
Watcharabutsarakham et al. | Leaf classification using structure features and support vector machines | |
CN106845489B (en) | SAR image target feature extraction method based on improved Krawtchouk moment | |
CN115049889A (en) | Storage medium and inference method | |
Rekik et al. | A k-means clustering algorithm initialization for unsupervised statistical satellite image segmentation | |
Hu et al. | Scale-invariant salient edge detection | |
CN113139077B (en) | Method, device, terminal and storage medium for identifying ship identity | |
Budiman et al. | Image feature extraction of numbers and letters using matrix segmentation |
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 |