CN111325094A - High-resolution range profile-based ship type identification method and system - Google Patents
High-resolution range profile-based ship type identification method and system Download PDFInfo
- Publication number
- CN111325094A CN111325094A CN202010046491.9A CN202010046491A CN111325094A CN 111325094 A CN111325094 A CN 111325094A CN 202010046491 A CN202010046491 A CN 202010046491A CN 111325094 A CN111325094 A CN 111325094A
- Authority
- CN
- China
- Prior art keywords
- loss function
- neural network
- network model
- ship
- ship type
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000003062 neural network model Methods 0.000 claims abstract description 81
- 238000000605 extraction Methods 0.000 claims abstract description 5
- 239000011159 matrix material Substances 0.000 claims description 18
- 238000013528 artificial neural network Methods 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 4
- 239000013598 vector Substances 0.000 description 4
- 230000008569 process Effects 0.000 description 3
- 230000002776 aggregation Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000002592 echocardiography Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
Images
Classifications
-
- 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
- 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
- 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
-
- 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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
-
- 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
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- 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
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Remote Sensing (AREA)
- Astronomy & Astrophysics (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Networks & Wireless Communication (AREA)
- Radar, Positioning & Navigation (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a ship type identification method and system based on a high-resolution range profile. The method comprises the steps of obtaining high-resolution distance images of ships of different ship types in a database; each ship of the ship type corresponds to a plurality of high-resolution range profiles; performing feature extraction on the high-resolution range profile to obtain a feature set; acquiring a neural network model; the neural network model takes the characteristic set as input and takes the ship type as output; training the neural network model according to a central loss function and a Softmax loss function; acquiring a high-resolution range profile of a ship to be identified; and identifying the ship type of the ship to be identified by using the trained neural network model. The ship type identification method and system based on the high-resolution range profile can improve the accuracy of classification and identification of the ship type.
Description
Technical Field
The invention relates to the field of ship type identification, in particular to a ship type identification method and system based on a high-resolution range profile.
Background
The radar target identification is to judge the type of a target by using a radar echo signal of the target. Broadband radars typically operate in the optical region, where the target can be viewed as being made up of a large number of scattering points of varying intensity. The High-resolution range profile (HRRP) is composed of a vector sum of echoes of scattering points on a ship acquired by a broadband radar signal. The method reflects the distribution condition of scattering points on the ship along the radar sight line, contains important structural characteristics of the ship, and is widely applied to the field of identification of the type of the radar ship. The ship type is identified by acquiring a high-resolution range profile of the ship, extracting the characteristics of the high-resolution range profile and constructing a neural network.
The final classification of the target features is done in the prior art by using a loss function. The loss function is used to measure the difference between the predicted value and the true value, and is generally denoted by L (y _, y), where y _ denotes the predicted value and y denotes the true value. For the Softmax loss function, the magnitude of the loss value is related to the modulo length of the feature. When the Softmax loss function classifies a certain sample correctly, Softmax continuous optimization can promote characteristic modular length to be continuously increased, but the characteristic modular length cannot be further improved by the change, and the generalization capability of the model is improved, because the ideal condition in target classification is to make the characteristics as close as possible, namely the Softmax loss function has the characteristics of easiness in optimization and quick convergence, and the Softmax loss function is generally used as a loss function in a multi-classification convolutional neural network. Softmax Loss is formally the sum of the Softmax function and the cross entropy Loss, and the classification of different classes is guaranteed to be correct by letting all target classes have the maximum log-likelihood in probability space. However, Softmax loss functions can cause the iterative training of the network, and Softmax only distinguishes different classes of features and does not try to separate the features more obviously. In the ship type classification and identification task, the characteristics processed by Softmax can divide the hyperspace or the hyperspace into corresponding blocks according to the number of the types of ships, the characteristics of Softmax determine that the characteristics of Softmax cannot guarantee the cluster aggregation and the cluster separation of the characteristics of various types, and then the trained network model cannot accurately classify and identify the types of the ships. And the center loss function (Centerlos, CL) constructs a class center for the characteristics of each class of targets, penalizes the target characteristics far away from the class center to promote the intra-class distance of the target characteristics to be more compact, and achieves the effects of reducing the intra-class distance and increasing the inter-class distance. But the central loss function is not suitable for direct feature classification because computing the features of all samples consumes huge computing resources when the number of training data sets is large.
Disclosure of Invention
The invention aims to provide a ship type identification method and system based on a high-resolution range profile, which can improve the accuracy of ship type classification identification.
In order to achieve the purpose, the invention provides the following scheme:
a ship type identification method based on high-resolution range profile comprises the following steps:
acquiring high-resolution distance images of ships of different ship types in a database; each ship of the ship type corresponds to a plurality of high-resolution range profiles;
performing feature extraction on the high-resolution range profile to obtain a feature set;
acquiring a neural network model; the neural network model takes the characteristic set as input and takes the ship type as output;
training the neural network model according to a central loss function and a Softmax loss function;
acquiring a high-resolution range profile of a ship to be identified;
and identifying the ship type of the ship to be identified by using the trained neural network model.
Optionally, the training the neural network model according to the central loss function and the Softmax loss function specifically includes:
using formulasDetermining a joint loss function; l isAMSCAs a joint loss function, LAMSAs a function of Softmax loss, LCAs a function of the central loss, λ is LAMSCMiddle LCThe weight of the weight to be occupied by the weight,is the ith characteristic and belongs to the yiEach category, d is the dimension of the feature;weight matrix for last full connection layerColumn j of (1);the weighted result of the ith sample target feature is obtained; m is the number of high-resolution range images of ships of different ship types in the database; n is the number of ship types; s is a scaling factor for scaling the cosine value, mu is an integer greater than 1, and is used for controlling the size of the corner edge between the features;is the yiA characteristic center of individual ship type;
and training the neural network model according to the joint loss function.
Optionally, the training the neural network model according to the central loss function and the Softmax loss function further includes:
Optionally, the training the neural network model according to the joint loss function specifically includes:
obtaining a deviation value of the combined loss function during the t iteration;
updating a full-connection matrix, convolution kernel parameters and a characteristic center of each ship type of the joint loss function in the neural network according to the deviation value;
calculating a deviation value of the updated joint loss function according to the updated neural network model and the updated characteristic center of each ship type of the joint loss function;
judging whether the deviation value of the updated joint loss function is larger than a set threshold value or not;
if the deviation value of the updated joint loss function is larger than the set threshold value, returning to the step of updating the full connection matrix, the convolution kernel parameters and the characteristic center of each ship type of the joint loss function in the neural network according to the deviation value;
and if the deviation value of the updated joint loss function is not greater than the set threshold, taking the currently updated neural network model as the trained neural network model.
A system for vessel type identification based on high resolution range profile, comprising:
the first acquisition module is used for acquiring high-resolution range images of ships of different ship types in the database; each ship of the ship type corresponds to a plurality of high-resolution range profiles;
the characteristic set determining module is used for extracting characteristics of the high-resolution range profile to obtain a characteristic set;
the neural network model acquisition module is used for acquiring a neural network model; the neural network model takes the characteristic set as input and takes the ship type as output;
the neural network model training module is used for training the neural network model according to a central loss function and a Softmax loss function;
the second acquisition module is used for acquiring a high-resolution range profile of the ship to be identified;
and the ship type identification module is used for identifying the ship type of the ship to be identified by using the trained neural network model.
Optionally, the neural network model training module specifically includes:
a joint loss function determination unit for utilizing the formulaDetermining a joint loss function; l isAMSCAs a joint loss function, LAMSAs a function of Softmax loss, LCAs a function of the central loss, λ is LAMSCMiddle LCThe weight of the weight to be occupied by the weight,is the ith characteristic and belongs to the yiEach category, d is the dimension of the feature;weight matrix for last full connection layerColumn j of (1);the weighted result of the ith sample target feature is obtained; m is the number of high-resolution range images of ships of different ship types in the database; n is the number of ship types; s is a scaling factor for scaling cosine values, and μ is an integer greater than 1 for controlling the angular edges between featuresThe size of the rim;is the yiA characteristic center of individual ship type;
and the neural network model training unit is used for training the neural network model according to the joint loss function.
Optionally, the method further includes:
a central loss function determination module for utilizing the formulaDetermining a central loss function;
a Softmax loss function determination module for utilizing the formulaA Softmax loss function is determined.
Optionally, the neural network model training unit specifically includes:
the first deviation value calculation operator unit is used for calculating the deviation value of the joint loss function when the tth iteration is obtained;
the updating subunit is used for updating the full-connection matrix, the convolution kernel parameters and the characteristic center of each ship type of the joint loss function in the neural network according to the deviation value;
the deviation value updating subunit is used for calculating the deviation value of the updated joint loss function according to the updated neural network model and the updated characteristic center of each ship type of the joint loss function;
a judging subunit, configured to judge whether a deviation value of the updated joint loss function is greater than a set threshold;
a continuous updating subunit, configured to, if the deviation value of the updated joint loss function is greater than the set threshold, return to the step of updating the full-connection matrix, the convolution kernel parameters, and the feature center of each ship type of the joint loss function in the neural network according to the deviation value;
and the trained neural network model determining subunit is used for taking the currently updated neural network model as the trained neural network model if the deviation value of the updated joint loss function is not greater than the set threshold value.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
according to the ship type identification method and system based on the high-resolution range profile, the central loss function and the Softmax loss function are combined with the training neural network model, the fact that the Softmax loss function increases the inter-class difference of different target characteristics through characteristic boundary constraint is considered, the fact that the central loss function reduces the intra-class difference of the different target characteristics through central clustering is considered, and the problem that the common Softmax only distinguishes the characteristics of different classes and cannot separate the characteristics is solved. The method has the advantages that the characteristics of intra-class aggregation and inter-class dispersion are met, and meanwhile, the accuracy of classification and identification of the types of ships is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
Fig. 1 is a schematic flow chart of a ship type identification method based on a high-resolution range profile according to the present invention;
fig. 2 is a schematic structural diagram of a ship type identification system based on a high-resolution range profile provided by the present invention.
Detailed Description
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 derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a ship type identification method and system based on a high-resolution range profile, which can improve the accuracy of ship type classification identification.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Fig. 1 is a schematic flow chart of a method for identifying a ship type based on a high-resolution range profile, as shown in fig. 1, the method for identifying a ship type based on a high-resolution range profile includes:
s101, acquiring high-resolution range images of ships of different ship types in a database; each ship of the ship type corresponds to a plurality of the high-resolution range images.
And S102, performing feature extraction on the high-resolution range profile to obtain a feature set.
S103, acquiring a neural network model; the neural network model takes the characteristic set as input and takes the ship type as output.
And S104, training the neural network model according to the central loss function and the Softmax loss function.
Using formulasA central loss function is determined. According to the formulaAnd formulaAnd updating the central loss function. When y isiFor the jth ship type, δ (-) is equal to 1 when identifying correct, otherwise δ (-) is equal to 0.
Using formulasA Softmax loss function is determined. The specific determination process of the Softmax loss function is as follows:
using formulasDetermining a first loss function; by adding multiplicative constraints, larger feature boundaries and recognition performance are obtained. The loss function A-Softmax is normalized by the weight vector (to W |)i| becomes 1) and willBecome intoThe form enhances the boundary constraint. The A-Softmax calculation formula is as follows:
where ψ (θ) is generally defined as a piecewise function as follows:
mu is typically an integer greater than 1 and is used to control the size of the corner edges between features. When mu is 1, then LASBecomes the conventional Softmax loss function. Mu is taken as a parameter closely related to the classification boundary of the features, the larger the value of mu is, the stronger the angle edge constraint is, the more difficult the model is to be converged during training, the model can be converged only by using various training skills, and certain difficulty exists in application. Further, the form of multiplying θ in ψ (θ) by the boundary constraint μ is converted to a more concise form of subtraction, i.e., ψ (θ) ═ cos θ - μ. This form is more versatile and simple to visualize than the a-Softmax loss function. In the actual operation process, the input after feature and weight normalization becomes:
therefore, only Ψ (f) ═ f — μ needs to be calculated during the forward propagation. And no extra calculation is needed in the back propagation process because Ψ' (f) ═ 1, which is easier to implement than the above formula.
The feature similarity can be calculated in two forms of Euclidean distance and cosine distance. Euclidean distance: the smaller the distance between the characteristic points is, the higher the similarity of the vectors is; cosine distance: the smaller the included angle between feature points, the higher the similarity of vectors. The cosine layer is constructed using the cosine as the similarity between two features, while the weights and normalization of the features are done. Introducing a hyper-parameter s as a scale factor to scale the cosine value, and obtaining a Softmax loss function as follows:
using formulasA joint loss function is determined. L isAMSCAs a joint loss function, LAMSAs a function of Softmax loss, LCAs a function of the central loss, λ is LAMSCMiddle LCThe weight of the weight to be occupied by the weight,is the ith characteristic and belongs to the yiEach category, d is the dimension of the feature;weight matrix for last full connection layerColumn j of (1);weighted node for ith sample target featureFruit; m is the number of high-resolution range images of ships of different ship types in the database; n is the number of ship types; s is a scaling factor for scaling the cosine value, mu is an integer greater than 1, and is used for controlling the size of the corner edge between the features;is the yiCharacteristic centers for individual ship types.
And training the neural network model according to the joint loss function.
Using formulasAndupdating the full-connection matrix W and the convolution kernel parameter theta in the neural networkcAnd a characteristic center c for each ship type of the joint loss functionj。
And calculating the deviation value of the updated joint loss function according to the updated neural network model and the updated characteristic center of each ship type of the joint loss function.
And judging whether the deviation value of the updated joint loss function is larger than a set threshold value.
And if the deviation value of the updated joint loss function is larger than the set threshold value, returning to the step of updating the full-connection matrix, the convolution kernel parameters and the characteristic center of each ship type of the joint loss function in the neural network according to the deviation value.
And if the deviation value of the updated joint loss function is not greater than the set threshold, taking the currently updated neural network model as the trained neural network model.
And S105, acquiring a high-resolution range image of the ship to be identified.
And S106, identifying the ship type of the ship to be identified by using the trained neural network model.
Fig. 2 is a schematic structural diagram of a ship type identification system based on a high-resolution range profile, as shown in fig. 2, the ship type identification system based on a high-resolution range profile provided in the present invention includes: the system comprises a first acquisition module 201, a feature set determination module 202, a neural network model acquisition module 203, a neural network model training module 204, a second acquisition module 205 and a ship type identification module 206.
The first obtaining module 201 is configured to obtain high-resolution range images of ships of different ship types in the database; each ship of the ship type corresponds to a plurality of the high-resolution range images.
The feature set determining module 202 is configured to perform feature extraction on the high-resolution range profile to obtain a feature set.
The neural network model obtaining module 203 is used for obtaining a neural network model; the neural network model takes the characteristic set as input and takes the ship type as output.
The neural network model training module 204 is configured to train the neural network model according to a central loss function and a Softmax loss function.
The second acquisition module 205 is configured to acquire a high-resolution range image of the ship to be identified.
The ship type identification module 206 is configured to identify the ship type of the ship to be identified by using the trained neural network model.
The neural network model training module 204 specifically includes: a joint loss function determination unit and a neural network model training unit.
A joint loss function determination unit for utilizing the formulaDetermining a joint loss function; l isAMSCAs a joint loss function, LAMSAs a function of Softmax loss, LCAs a function of the central loss, λ is LAMSCMiddle LCThe weight of the weight to be occupied by the weight,is the ith characteristic and belongs to the yiEach category, d is the dimension of the feature;weight matrix for last full connection layerColumn j of (1);the weighted result of the ith sample target feature is obtained; m is the number of high-resolution range images of ships of different ship types in the database; n is the number of ship types; s is a scaling factor for scaling the cosine value, mu is an integer greater than 1, and is used for controlling the size of the corner edge between the features;is the yiCharacteristic centers for individual ship types.
And the neural network model training unit is used for training the neural network model according to the joint loss function.
The ship type identification system based on the high-resolution range profile further comprises: a central loss function determination module and a Softmax loss function determination module.
A central loss function determination module for utilizing the formulaA central loss function is determined.
The Softmax loss function determining module is used for utilizing a formulaA Softmax loss function is determined.
The neural network model training unit specifically comprises: the first deviation value calculating subunit, the updating subunit, the deviation value updating subunit, the judging subunit, the continuous updating subunit and the trained neural network model determining subunit.
And the first deviation value calculating unit is used for calculating the deviation value of the joint loss function when the t iteration is obtained.
And the updating unit is used for updating the full-connection matrix, the convolution kernel parameters and the characteristic center of each ship type of the joint loss function in the neural network according to the deviation value.
And the deviation value updating subunit is used for calculating the deviation value of the updated joint loss function according to the updated neural network model and the updated characteristic center of each ship type of the joint loss function.
And the judgment subunit is used for judging whether the deviation value of the updated joint loss function is greater than a set threshold value.
And the continuous updating subunit is configured to return to the step of updating the full connection matrix, the convolution kernel parameter, and the feature center of each ship type of the joint loss function in the neural network according to the deviation value if the deviation value of the updated joint loss function is greater than the set threshold value.
And the trained neural network model determining subunit is used for taking the currently updated neural network model as the trained neural network model if the deviation value of the updated joint loss function is not greater than the set threshold value.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (8)
1. A ship type identification method based on high-resolution range profile is characterized by comprising the following steps:
acquiring high-resolution distance images of ships of different ship types in a database; each ship of the ship type corresponds to a plurality of high-resolution range profiles;
performing feature extraction on the high-resolution range profile to obtain a feature set;
acquiring a neural network model; the neural network model takes the characteristic set as input and takes the ship type as output;
training the neural network model according to a central loss function and a Softmax loss function;
acquiring a high-resolution range profile of a ship to be identified;
and identifying the ship type of the ship to be identified by using the trained neural network model.
2. The method for identifying a ship type based on a high-resolution range profile as claimed in claim 1, wherein the training of the neural network model according to a central loss function and a Softmax loss function specifically comprises:
using formulasDetermining a joint loss function; l isAMSCAs a joint loss function, LAMSAs a function of Softmax loss, LCAs a function of the central loss, λ is LAMSCMiddle LCThe weight of the weight to be occupied by the weight,is the ith characteristic and belongs to the yiA class, d is special(ii) a characterized dimension;weight matrix for last full connection layerColumn j of (1);the weighted result of the ith sample target feature is obtained; m is the number of high-resolution range images of ships of different ship types in the database; n is the number of ship types; s is a scaling factor for scaling the cosine value, mu is an integer greater than 1, and is used for controlling the size of the corner edge between the features;is the yiA characteristic center of individual ship type;
and training the neural network model according to the joint loss function.
3. The method for identifying ship types based on high-resolution range profiles as claimed in claim 2, wherein the training of the neural network model according to the central loss function and the Softmax loss function further comprises:
4. The method according to claim 3, wherein the training of the neural network model according to the joint loss function specifically includes:
obtaining a deviation value of the combined loss function during the t iteration;
updating a full-connection matrix, convolution kernel parameters and a characteristic center of each ship type of the joint loss function in the neural network according to the deviation value;
calculating a deviation value of the updated joint loss function according to the updated neural network model and the updated characteristic center of each ship type of the joint loss function;
judging whether the deviation value of the updated joint loss function is larger than a set threshold value or not;
if the deviation value of the updated joint loss function is larger than the set threshold value, returning to the step of updating the full connection matrix, the convolution kernel parameters and the characteristic center of each ship type of the joint loss function in the neural network according to the deviation value;
and if the deviation value of the updated joint loss function is not greater than the set threshold, taking the currently updated neural network model as the trained neural network model.
5. A system for identifying a ship type based on a high-resolution range profile, comprising:
the first acquisition module is used for acquiring high-resolution range images of ships of different ship types in the database; each ship of the ship type corresponds to a plurality of high-resolution range profiles;
the characteristic set determining module is used for extracting characteristics of the high-resolution range profile to obtain a characteristic set;
the neural network model acquisition module is used for acquiring a neural network model; the neural network model takes the characteristic set as input and takes the ship type as output;
the neural network model training module is used for training the neural network model according to a central loss function and a Softmax loss function;
the second acquisition module is used for acquiring a high-resolution range profile of the ship to be identified;
and the ship type identification module is used for identifying the ship type of the ship to be identified by using the trained neural network model.
6. The system for identifying ship types based on high-resolution range profiles as claimed in claim 5, wherein the neural network model training module specifically comprises:
a joint loss function determination unit for utilizing the formulaDetermining a joint loss function; l isAMSCAs a joint loss function, LAMSAs a function of Softmax loss, LCAs a function of the central loss, λ is LAMSCMiddle LCThe weight of the weight to be occupied by the weight,is the ith characteristic and belongs to the yiEach category, d is the dimension of the feature;weight matrix for last full connection layerColumn j of (1);the weighted result of the ith sample target feature is obtained; m is the number of high-resolution range images of ships of different ship types in the database; n is the number of ship types; s is a scaling factor for scaling the cosine value, mu is an integer greater than 1, and is used for controlling the size of the corner edge between the features;is the yiA characteristic center of individual ship type;
and the neural network model training unit is used for training the neural network model according to the joint loss function.
8. The system according to claim 7, wherein the neural network model training unit specifically comprises:
the first deviation value calculation operator unit is used for calculating the deviation value of the joint loss function when the tth iteration is obtained;
the updating subunit is used for updating the full-connection matrix, the convolution kernel parameters and the characteristic center of each ship type of the joint loss function in the neural network according to the deviation value;
the deviation value updating subunit is used for calculating the deviation value of the updated joint loss function according to the updated neural network model and the updated characteristic center of each ship type of the joint loss function;
a judging subunit, configured to judge whether a deviation value of the updated joint loss function is greater than a set threshold;
a continuous updating subunit, configured to, if the deviation value of the updated joint loss function is greater than the set threshold, return to the step of updating the full-connection matrix, the convolution kernel parameters, and the feature center of each ship type of the joint loss function in the neural network according to the deviation value;
and the trained neural network model determining subunit is used for taking the currently updated neural network model as the trained neural network model if the deviation value of the updated joint loss function is not greater than the set threshold value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010046491.9A CN111325094A (en) | 2020-01-16 | 2020-01-16 | High-resolution range profile-based ship type identification method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010046491.9A CN111325094A (en) | 2020-01-16 | 2020-01-16 | High-resolution range profile-based ship type identification method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111325094A true CN111325094A (en) | 2020-06-23 |
Family
ID=71172539
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010046491.9A Pending CN111325094A (en) | 2020-01-16 | 2020-01-16 | High-resolution range profile-based ship type identification method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111325094A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113406623A (en) * | 2021-05-07 | 2021-09-17 | 中山大学 | Target identification method, device and medium based on radar high-resolution range profile |
CN114898222A (en) * | 2022-04-21 | 2022-08-12 | 中国人民解放军91977部队 | Ship target track identification method and device |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108446689A (en) * | 2018-05-30 | 2018-08-24 | 南京开为网络科技有限公司 | A kind of face identification method |
CN109033938A (en) * | 2018-06-01 | 2018-12-18 | 上海阅面网络科技有限公司 | A kind of face identification method based on ga s safety degree Fusion Features |
CN109085918A (en) * | 2018-06-28 | 2018-12-25 | 天津大学 | Acupuncture needling manipulation training method based on myoelectricity |
CN110263603A (en) * | 2018-05-14 | 2019-09-20 | 桂林远望智能通信科技有限公司 | Face identification method and device based on center loss and residual error visual simulation network |
CN110598801A (en) * | 2019-09-24 | 2019-12-20 | 东北大学 | Vehicle type recognition method based on convolutional neural network |
-
2020
- 2020-01-16 CN CN202010046491.9A patent/CN111325094A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110263603A (en) * | 2018-05-14 | 2019-09-20 | 桂林远望智能通信科技有限公司 | Face identification method and device based on center loss and residual error visual simulation network |
CN108446689A (en) * | 2018-05-30 | 2018-08-24 | 南京开为网络科技有限公司 | A kind of face identification method |
CN109033938A (en) * | 2018-06-01 | 2018-12-18 | 上海阅面网络科技有限公司 | A kind of face identification method based on ga s safety degree Fusion Features |
CN109085918A (en) * | 2018-06-28 | 2018-12-25 | 天津大学 | Acupuncture needling manipulation training method based on myoelectricity |
CN110598801A (en) * | 2019-09-24 | 2019-12-20 | 东北大学 | Vehicle type recognition method based on convolutional neural network |
Non-Patent Citations (5)
Title |
---|
FENG WANG 等: "Additive Margin Softmax for Face Verification", 《IEEE SIGNAL PROCESSING LETTERS》 * |
MINGCHAO JIANG 等: "Additive Margin Softmax with Center Loss for Face Recognition", 《ICVIP 2018》 * |
余东行 等: "联合显著性特征与卷积神经网络的遥感影像舰船检测", 《中国图象图形学报》 * |
余成波 等: "中心损失与Softmax损失联合监督下的人脸识别", 《重庆大学学报》 * |
郭晨 等: "基于深度多尺度一维卷积神经网络的雷达舰船目标识别", 《电子与信息学报》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113406623A (en) * | 2021-05-07 | 2021-09-17 | 中山大学 | Target identification method, device and medium based on radar high-resolution range profile |
CN114898222A (en) * | 2022-04-21 | 2022-08-12 | 中国人民解放军91977部队 | Ship target track identification method and device |
CN114898222B (en) * | 2022-04-21 | 2024-01-02 | 中国人民解放军91977部队 | Ship target track identification method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111507335B (en) | Method and device for automatically labeling training images used for deep learning network | |
CN109919108B (en) | Remote sensing image rapid target detection method based on deep hash auxiliary network | |
CN110472483B (en) | SAR image-oriented small sample semantic feature enhancement method and device | |
US11670071B2 (en) | Fine-grained image recognition | |
CN111160407B (en) | Deep learning target detection method and system | |
CN109063719B (en) | Image classification method combining structure similarity and class information | |
CN111160249A (en) | Multi-class target detection method of optical remote sensing image based on cross-scale feature fusion | |
CN111428557A (en) | Method and device for automatically checking handwritten signature based on neural network model | |
CN113095333B (en) | Unsupervised feature point detection method and unsupervised feature point detection device | |
Liu et al. | An improved InceptionV3 network for obscured ship classification in remote sensing images | |
CN110569738A (en) | natural scene text detection method, equipment and medium based on dense connection network | |
CN113743417B (en) | Semantic segmentation method and semantic segmentation device | |
CN117409190B (en) | Real-time infrared image target detection method, device, equipment and storage medium | |
CN112529068B (en) | Multi-view image classification method, system, computer equipment and storage medium | |
CN110852327A (en) | Image processing method, image processing device, electronic equipment and storage medium | |
CN111325094A (en) | High-resolution range profile-based ship type identification method and system | |
CN111738319B (en) | Clustering result evaluation method and device based on large-scale samples | |
CN114926693A (en) | SAR image small sample identification method and device based on weighted distance | |
CN111079807B (en) | Ground object classification method and device | |
CN107564013B (en) | Scene segmentation correction method and system fusing local information | |
CN117437423A (en) | Weak supervision medical image segmentation method and device based on SAM collaborative learning and cross-layer feature aggregation enhancement | |
US20230401670A1 (en) | Multi-scale autoencoder generation method, electronic device and readable storage medium | |
CN113011376B (en) | Marine ship remote sensing classification method and device, computer equipment and storage medium | |
CN115331021A (en) | Dynamic feature extraction and description method based on multilayer feature self-difference fusion | |
CN114612802A (en) | System and method for classifying fine granularity of ship target based on MBCNN |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200623 |