CN115205680A - Radar target SAR image joint detection and identification method based on significance migration - Google Patents
Radar target SAR image joint detection and identification method based on significance migration Download PDFInfo
- Publication number
- CN115205680A CN115205680A CN202210827227.8A CN202210827227A CN115205680A CN 115205680 A CN115205680 A CN 115205680A CN 202210827227 A CN202210827227 A CN 202210827227A CN 115205680 A CN115205680 A CN 115205680A
- Authority
- CN
- China
- Prior art keywords
- target
- deep learning
- learning model
- training set
- image
- 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
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
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/34—Smoothing or thinning of the pattern; Morphological operations; Skeletonisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/772—Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The application relates to a radar target SAR image joint detection and identification method based on significance migration. The method comprises the following steps: preprocessing a radar target two-dimensional image to obtain an SAR gray level image; dividing the SAR gray level image into a training set and a test set, and performing random data enhancement and normalization processing on the training set to obtain a normalized training set; constructing a deep learning model; the deep learning model comprises a double-layer convolution, a capsule network, significance detection and a feature layer mask; training the deep learning model by using the normalized training set to obtain a trained deep learning model; and identifying the test set according to the trained deep learning model to obtain an identification result. By adopting the method, the radar target image detection accuracy can be improved.
Description
Technical Field
The application relates to the technical field of radar target identification, in particular to a radar target SAR image joint detection identification method and device based on significance migration, computer equipment and a storage medium.
Background
The synthetic space radar (SAR) has important application value in the aspect of remote sensing detection as an active imaging device. Because it can work all day long and all weather, it has been widely used in military and civil fields, such as automatic target identification, battlefield situation perception and earth remote sensing measurement and control. Due to a special imaging mechanism of the SAR, the SAR image comprises target clutter and background clutter. Therefore, the basic problem in the field of automatic target recognition of SAR images is to detect and recognize targets in background clutter.
The existing method generally decomposes the problem into two-stage tasks, the detection task separates a target and a background clutter from an SAR image, and the identification task determines the type of the target. At present, a deep learning technology widely applied to SAR automatic target recognition usually ignores or simplifies a detection task, and a target cutting region or a threshold filtering method is adopted to remove background clutter influence. The clipping operation reserves background clutter in the slice, and the threshold filtering method depends on parameter selection, does not integrally consider detection and identification tasks, and has low accuracy rate of target detection and identification in different scenes.
Disclosure of Invention
Therefore, in order to solve the above technical problems, it is necessary to provide a radar target SAR image joint detection and identification method, an apparatus, a computer device, and a storage medium based on saliency migration, which can improve the detection and identification accuracy of a radar target SAR image in different scenes.
A radar target SAR image joint detection and identification method based on significance migration comprises the following steps:
acquiring a two-dimensional image of a radar target to be identified;
preprocessing a radar target two-dimensional image to obtain an SAR gray level image;
dividing the SAR gray level image into a training set and a test set, and performing random data enhancement and normalization processing on the training set to obtain a normalized training set;
constructing a deep learning model; the deep learning model comprises a double-layer convolution, a capsule network, significance detection and a feature layer mask; the deep learning model is of a U-shaped structure;
training the deep learning model by using the normalized training set to obtain a trained deep learning model;
and identifying the test set according to the trained deep learning model to obtain an identification result.
In one embodiment, the training of the deep learning model by using the normalized training set to obtain the trained deep learning model includes:
obtaining a plurality of different training set samples from the normalized training set;
extracting target information of different training set samples by using double-layer convolution to obtain multi-scale target characteristics;
performing feature aggregation on the multi-scale target features according to the U-shaped structure of the deep learning model to obtain enhanced multi-scale target features;
migrating target importance information from an optical pre-training model by using the significance map as a pseudo label of the significance detection to obtain a target key area;
performing target and background separation on the enhanced multi-scale target features according to the feature layer mask to obtain target features;
inputting the target characteristics into the capsule network for characteristic conversion to obtain target space characteristics;
and training the deep learning model by using the target key region, the target spatial characteristics and the preset loss function to obtain the trained deep learning model.
In one embodiment, the process of obtaining the saliency map comprises:
training the VGG16 model pre-trained on ImageNet under the MSTAR standard working condition data set to obtain a trained VGG16 model;
and extracting various significance maps from the trained VGG16 model according to the Guided Grad-Cam.
In one embodiment, the preset penalty function comprises interval penalty of the capsule network, significance detection two-class cross entropy penalty and target area mask L 1 And (4) regularizing.
In one embodiment, the preset loss function is
L＝L m +α·L S +β·L 1
L j ＝T j max(0,m + -||v j ||) 2 +
λ(1-T j )max(0,||v j ||-m - ) 2
Wherein m is + =0.9 true positive lower bound, m - =0.1 for true negative upper bound, λ =0.5 for scaling factor, L m Represents the spacing loss, L, of the capsule network S Represents significance detection two-class cross entropy loss, L 1 1-norm regularization, L, representing a mask of a target area j Denotes the loss of class j, T j Representing an indicator function, T j 1 only if it is the j-th class, the hyper-parameter α is set to 1e-5 and β is set to 5e-6 to control the mask sparsity.
In one embodiment, preprocessing a radar target two-dimensional image to obtain an SAR grayscale image includes:
performing linear transformation and gray level enhancement on the radar target two-dimensional image to obtain an SAR gray level image;
the calculation process of the linear transformation is
Wherein x represents the original radar target two-dimensional image pixel input value, y represents the output image pixel value, and x max Representing the maximum value of the pixels, x, of the two-dimensional image of the original radar target min Representing the minimum value of the pixels of the two-dimensional image of the original radar target;
the calculation process of the gray enhancement is
Wherein I represents the gray value of the radar target two-dimensional image, O represents the gray value of the SAR gray image, I min_count Gray value, I, representing radar target two-dimensional image with minimum occurrence max_count And representing the gray value with the largest occurrence number of the two-dimensional image of the radar target.
In one embodiment, the random data enhancement and normalization processing on the training set to obtain a normalized training set includes:
carrying out random data enhancement and normalization processing on the training set according to image rotation, gaussian white noise disturbance, random uniform noise replacement and a preset random data enhancement method implementation rate to obtain a normalized training set; the implementation rate of each method in random data enhancement is respectively 0.3.
A radar target SAR image joint detection and identification device based on significance migration, the device comprises:
the preprocessing module is used for acquiring a two-dimensional image of a radar target to be identified; preprocessing a radar target two-dimensional image to obtain an SAR gray level image;
the random enhancement and normalization module is used for dividing the SAR gray level image into a training set and a test set, and performing random data enhancement and normalization processing on the training set to obtain a normalized training set;
the model building and training module is used for building a deep learning model; the deep learning model comprises double-layer convolution, a capsule network, significance detection and a feature layer mask; the deep learning model is of a U-shaped structure; training the deep learning model by using the normalized training set to obtain a trained deep learning model;
and the recognition module is used for recognizing the test set according to the trained deep learning model to obtain a recognition result.
A computer device comprising a memory and a processor, the memory storing a computer program, the processor implementing the following steps when executing the computer program:
acquiring a two-dimensional image of a radar target to be identified;
preprocessing a radar target two-dimensional image to obtain an SAR gray level image;
dividing the SAR gray level image into a training set and a test set, and performing random data enhancement and normalization processing on the training set to obtain a normalized training set;
constructing a deep learning model; the deep learning model comprises double-layer convolution, a capsule network, significance detection and a feature layer mask; the deep learning model is of a U-shaped structure;
training the deep learning model by using the normalized training set to obtain a trained deep learning model;
and identifying the test set according to the trained deep learning model to obtain an identification result.
A computer-readable storage medium, on which a computer program is stored which, when executed by a processor, carries out the steps of:
acquiring a two-dimensional image of a radar target to be identified;
preprocessing a radar target two-dimensional image to obtain an SAR gray level image;
dividing the SAR gray level image into a training set and a test set, and performing random data enhancement and normalization processing on the training set to obtain a normalized training set;
constructing a deep learning model; the deep learning model comprises a double-layer convolution, a capsule network, significance detection and a feature layer mask; the deep learning model is of a U-shaped structure;
training the deep learning model by using the normalized training set to obtain a trained deep learning model;
and identifying the test set according to the trained deep learning model to obtain an identification result.
According to the radar target SAR image joint detection and identification method, device, computer equipment and storage medium based on significance migration, firstly, a radar target two-dimensional image is preprocessed to obtain an SAR gray level image, SAR image target details are enhanced, more accurate image characteristic information is extracted subsequently, the accuracy of image detection is further improved, the SAR gray level image is divided into a training set and a testing set, random data enhancement and normalization processing are carried out on the training set, the obtained normalized training set is used for constructing a deep learning model; the deep learning model comprises a double-layer convolution, a capsule network, significance detection and a feature layer mask, wherein the significance map of the pre-training model is used as a pseudo label of the significance detection, so that the key feature region information of a target is migrated, the feature layer mask is used for separating the target from a background, the significance detection and the capsule network recognition are integrally designed in the deep learning model to serve as a detection and recognition functional module, so that the deep learning model can effectively extract the spatial information of the target in a scene, the detection and recognition accuracy is improved, the deep learning model has good generalization performance, the position target in an unknown scene can be detected due to good recognition robustness in different scenes, the significance map of the pre-training model is used as the pseudo label of the significance detection, the target significance information is migrated, and the defects of long time consumption and high cost of manual labeling are avoided.
Drawings
Fig. 1 is a schematic flow chart of a radar target SAR image joint detection and identification method based on saliency migration in one embodiment;
FIG. 2 is a diagram of a deep learning model architecture in one embodiment;
FIG. 3 is a pseudo-label diagram of a saliency map for a saliency detection task in one embodiment;
FIG. 4 is a diagram illustrating MSTAR data set detection results under different scenarios in another embodiment;
FIG. 5 is a schematic diagram of a detection structure of a surface vessel under different scenarios in another embodiment;
FIG. 6 is a block diagram illustrating a structure of a radar target SAR image joint detection and identification device based on saliency migration in one embodiment;
FIG. 7 is a diagram illustrating an internal structure of a computer device according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
In one embodiment, as shown in fig. 1, a radar target SAR image joint detection and identification method based on saliency migration is provided, which includes the following steps:
102, acquiring a two-dimensional image of a radar target to be identified; and preprocessing the radar target two-dimensional image to obtain an SAR gray level image.
The method comprises the steps of carrying out data preprocessing methods such as linear transformation and gray level enhancement on an original synthetic aperture radar two-dimensional image of a radar target, converting the data preprocessing method into an SAR gray level image, enhancing target details of the SAR image, enabling more accurate image characteristic information to be extracted subsequently, further improving the accuracy of image detection, and dividing the processed SAR gray level image into a training set and a testing set for carrying out model training subsequently. The size of each SAR grayscale image is a two-dimensional matrix.
And step 104, dividing the SAR gray level image into a training set and a test set, and performing random data enhancement and normalization processing on the training set to obtain a normalized training set.
Random data enhancement is carried out on the training set, sample diversity is enhanced, and the normalized training sample is used as a training sample, so that the accuracy of model training can be improved. The random data enhancement method comprises image rotation, additive white Gaussian noise and uniform noise replacement. The probability of implementing image rotation, additive white gaussian noise and uniform noise is 0.3, 0.2 and 0.2 respectively. The image rotation rotates the original image by a small angle, the sample diversity is enhanced, and the rotation angle is uniform between [ -5,5 ]. The additive white Gaussian noise method is characterized in that the additive white Gaussian noise is added on the basis of an original image, the influence of different scene transformations on the intensity of scattering points of a target is simulated, the mean value is 0.1, the standard deviation is 0.1, and the amplitude is uniformly valued between [0.5,1.5 ]. The uniform noise replacement adopts uniformly distributed noise between [0,1], and the original image replacement proportion is between [0,0.05 ]. To avoid the sensitivity of the model to single pixel point changes.
106, constructing a deep learning model; the deep learning model comprises a double-layer convolution, a capsule network, significance detection and a feature layer mask; the deep learning model is of a U-shaped structure; and training the deep learning model by using the normalized training set to obtain the trained deep learning model.
The deep learning model is designed into a U-shaped structure, each layer of information of the front four layers of the U-shaped structure is used as multi-scale feature information to carry out feature aggregation, feature information of different scales of each layer describes structure information of different layers of a target, low-layer feature information represents information such as target texture and structure, high-layer feature information represents information such as target semantic category, and therefore the performance of a detection and identification function module of the integrated design of significance detection and capsule network identification is enhanced.
The deep learning model comprises a double-layer convolution, a capsule network, significance detection and a feature layer mask, wherein a significance map of a pre-training model is used as a pseudo label of the significance detection to migrate target key feature area information, the feature layer mask is used for separating a target from a background, the significance detection and the capsule network recognition are integrally designed in the deep learning model to serve as a detection and recognition function module, so that the deep learning model can effectively extract space information of the target in a scene, the accuracy of detection and recognition is improved, the deep learning model has good generalization performance, the position target in an unknown scene can be detected due to good recognition robustness in different scenes, the significance map of the pre-training model is used as the significance information of the pseudo label migration target of the significance detection, and the defects of long time consumption and high cost of manual labeling are avoided, the pseudo label map is shown in figure 2, the first line is an SAR gray level image of an MSTAR data set, and the significance map result corresponding to the second line is obtained.
And 108, identifying the test set according to the trained deep learning model to obtain an identification result.
In the radar target SAR image joint detection and identification method based on significance migration, firstly, a radar target two-dimensional image is preprocessed to obtain an SAR gray level image, SAR image target details are enhanced, more accurate image characteristic information is extracted subsequently, the accuracy of image detection is further improved, the SAR gray level image is divided into a training set and a testing set, random data enhancement and normalization processing are carried out on the training set, the obtained normalized training set is obtained, and a deep learning model is constructed; the deep learning model comprises a double-layer convolution, a capsule network, significance detection and a feature layer mask, wherein the significance graph of the pre-training model is used as a pseudo label of the significance detection to migrate the key feature region information of a target, the feature layer mask is used for mask separation of the target and a background, the significance detection and the capsule network recognition are integrally designed in the deep learning model to serve as a detection and recognition function module, so that the deep learning model can effectively extract the spatial information of the target in a scene, the detection and recognition accuracy is improved, the deep learning model has good generalization performance, the position target in an unknown scene can be detected due to good recognition robustness in different scenes, the significance graph of the pre-training model is used as significance detection pseudo label migration target significance information, and the defects of long time consumption and high cost of manual labeling are avoided.
In one embodiment, the training of the deep learning model by using the normalized training set to obtain the trained deep learning model includes:
obtaining a plurality of different training set samples from the normalized training set;
extracting target information of different training set samples by using double-layer convolution to obtain multi-scale target characteristics;
performing feature aggregation on the multi-scale target features according to the U-shaped structure of the deep learning model to obtain enhanced multi-scale target features;
migrating target importance information from an optical pre-training model by using the significance map as a pseudo label of the significance detection to obtain a target key area;
performing target and background separation on the enhanced multi-scale target features according to the feature layer mask to obtain target features;
inputting the target characteristics into the capsule network for characteristic conversion to obtain target space characteristics;
and training the deep learning model by using the target key region, the target spatial characteristics and the preset loss function to obtain the trained deep learning model.
In one embodiment, the process of obtaining the saliency map comprises:
training the VGG16 model pre-trained on ImageNet under the MSTAR standard working condition data set to obtain a trained VGG16 model;
and extracting various significance maps from the trained VGG16 model according to the Guided Grad-Cam.
In a specific embodiment, the various saliency maps are summed to eliminate class distinction and normalized to [0,1] to obtain the target area of emphasis.
In one embodiment, the preset penalty function includes interval penalty of the capsule network, significance detection two-class cross entropy penalty, and target area mask L 1 And (4) regularizing.
In one embodiment, the preset loss function is
L＝L m +α·L S +β·L 1
L j ＝T j max(0,m + -||v j ||) 2 +
λ(1-T j )max(0,||v j ||-m - ) 2
Wherein m is + =0.9 represents the lower limit of true positive, m - =0.1 for true negative upper bound, λ =0.5 for proportionality factor, L m Represents the spacing loss, L, of the capsule network S Represents the significance detection two-class cross entropy loss, L 1 1-norm regularization, L, representing a mask of a target area j Denotes the loss of class j, T j Representing an indicator function, T j Only if 1 is the jth class, the hyper-parameter α is set to 1e-5 and β is set to 5e-6 to control the mask sparsity.
In a specific embodiment, the deep learning model combining U-shape structure, saliency detection and feature layer mask, as shown in fig. 2, adapts to different size pictures by adaptively averaging the pooling layer such that its input image size is not fixed to 128 × 128, where the two-layer convolution uses two times the convolutional layer, the batch normalization layer structure and the ReLU activation function repeatedly, the kernel size of the convolutional layer is 5 × 5, the step size is 1 × 1, and the padding size is 2 × 2. The U-shaped structure shown in FIG. 2 utilizes information of different scales, and enhances the performance of the detection and identification module. The information of different scales uses the information of each layer of the front four layers of the U-shaped structure as multi-scale characteristic information.
Extracting a significance Map of a VGG16 pre-training model by using a Guided Grad-Cam, training the VGG16 pre-training model on ImageNet under a standard working condition (SOC) data set of MSTAR, extracting the significance Map of the trained VGG16 model based on the Guided Grad-Cam, and extracting various significance maps Map c Summing to obtain Map multi To eliminate class discrimination and normalize to [0,1]Summing the various classes of saliency maps as pseudo-labels is used to eliminate class distinction and normalize to [0,1]. The feature layer masking is performed on the intermediate feature layer to utilize the good generalization performance of the intermediate layer features, so that the method has good detection generalization performance on unknown scenes and targets. Identifying task using capsule networkAnd (5) processing the multi-scale features to obtain a recognition result. The initial capsule layer convolution kernel size of the capsule network is 5 x 5 with a step size of 2 x 2. The input capsule size of the routing capsule layer is 576 multiplied by 8 and the output capsule length is 8.
When model training is performed, the network training parameters are as follows: the training batch is 100, the batch training size is 64, the learning rate is 3e-4, the exponential decay rate is 0.98, and the optimizer uses the NAdam algorithm.
In one embodiment, preprocessing a radar target two-dimensional image to obtain an SAR grayscale image includes:
performing linear transformation and gray level enhancement on the radar target two-dimensional image to obtain an SAR gray level image;
the calculation process of the linear transformation is
Wherein x represents the original radar target two-dimensional image pixel input value, y represents the output image pixel value, and x max Representing the maximum value of the pixels, x, of the two-dimensional image of the original radar target min Representing the minimum value of the pixels of the two-dimensional image of the original radar target;
the calculation process of the gray enhancement is
Wherein I represents the gray value of the radar target two-dimensional image, O represents the gray value of the SAR gray image, I min_count Gray value, I, representing radar target two-dimensional image with minimum occurrence max_count And representing the gray value with the largest occurrence number of the two-dimensional image of the radar target.
In one embodiment, the random data enhancement and normalization processing on the training set to obtain a normalized training set includes:
carrying out random data enhancement and normalization processing on the training set according to image rotation, gaussian white noise disturbance, random uniform noise replacement and a preset random data enhancement method implementation rate to obtain a normalized training set; the implementation rate of each method in random data enhancement is respectively 0.3.
In one embodiment, a deep learning model is constructed based on the pytorch framework. As shown in table 1, tables are set for training sets and test sets in three different scenarios of the MSTAR data set, and measured data are radar targets in three different scenarios of the MSTAR data set. The algorithm performance is tested by repeating ten times of experiments to calculate the recognition rate and the variance. The comparison method selects three methods of A-ConvNet, MVGGNet and Extended conditional Capsule Network (ECCNet).
TABLE 1
In order to verify the generalization performance of the detection task, a multisource multiscale SAR ship slice data set is adopted for verification, and SAR images in the data set are derived from data of a domestic high-resolution No. 3 satellite and a European space agency Sentinel-1 satellite.
And testing the network performance by using 128 x 128SAR image slices under three different scenes of an MSTAR data set, wherein the scene 1 data is adopted in a training set, and the scenes 2 and 3 data are adopted in a testing set. The overall recognition rate (OA) and standard deviation (STD) of the recognition results of the other methods and the present invention on the test set are shown in table 2. The measured data shows that the method can effectively extract the spatial information of the target in the scene, remove the influence of background clutter in different scenes, and still have good identification performance when the training set and the test set are in different scenes.
TABLE 2
The detection result of the MSTAR dataset is shown in fig. 4, in which the upper line in the graph is the SAR grayscale image of the MSTAR dataset, and the lower line is the corresponding detection result of the present invention. The detection module can detect the target area from different background clutter in different scenes, and can effectively separate the target area from the background clutter.
In order to test the generalization performance of the detection module, the model trained on the MSTAR data set is used for detecting the SAR image of the sea surface ship, and the generalization performance of the detection module on unknown environment and unknown targets is verified. The size of the multisource multiscale SAR ship slice data set is 256 multiplied by 256, the detection result is shown in figure 5, the SAR gray level image of the multisource multiscale SAR ship slice data set is taken as the upper action, and the detection result of the invention is taken as the lower action.
According to experimental results, the significance map of the pre-training model is used as the pseudo label of the significance detection, the key feature region information of the target is migrated, and the feature layer target region mask is used for separating the target from the background, so that the model can effectively extract the spatial information of the target in the scene. The method of the invention has good identification performance in different scenes, and can detect the position target in unknown scenes.
It should be understood that, although the steps in the flowchart of fig. 1 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a portion of the steps in fig. 1 may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performance of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.
In one embodiment, as shown in fig. 6, a radar target SAR image joint detection and identification device based on significance migration is provided, including: a pre-processing module 602, a stochastic enhancement and normalization module 604, a model construction and training module 606, and a recognition module 608, wherein:
the preprocessing module 602 is configured to obtain a two-dimensional image of a radar target to be identified; preprocessing a radar target two-dimensional image to obtain an SAR gray level image;
a random enhancing and normalizing module 604, configured to divide the SAR grayscale image into a training set and a test set, and perform random data enhancing and normalizing processing on the training set to obtain a normalized training set;
a model construction and training module 606 for constructing a deep learning model; the deep learning model comprises a double-layer convolution, a capsule network, significance detection and a feature layer mask; the deep learning model is of a U-shaped structure; training the deep learning model by using the normalized training set to obtain a trained deep learning model;
and the recognition module 608 is configured to recognize the test set according to the trained deep learning model to obtain a recognition result.
In one embodiment, the model building and training module 606 is further configured to train the deep learning model by using the normalized training set, so as to obtain a trained deep learning model, including:
obtaining a plurality of different training set samples from the normalized training set; extracting target information of different training set samples by using double-layer convolution to obtain multi-scale target characteristics;
performing feature aggregation on the multi-scale target features according to the U-shaped structure of the deep learning model to obtain enhanced multi-scale target features;
migrating target importance information from an optical pre-training model by using the significance map as a pseudo label of the significance detection to obtain a target key area;
performing target and background separation on the enhanced multi-scale target features according to the feature layer mask to obtain target features;
inputting the target characteristics into the capsule network for characteristic conversion to obtain target space characteristics;
and training the deep learning model by using the target key region, the target spatial characteristics and the preset loss function to obtain the trained deep learning model.
In one embodiment, the process of obtaining the saliency map comprises:
training the VGG16 model pre-trained on ImageNet under the MSTAR standard working condition data set to obtain a trained VGG16 model;
and extracting various significance maps from the trained VGG16 model according to the Guided Grad-Cam.
In one embodiment, the preset penalty function comprises interval penalty of the capsule network, significance detection two-class cross entropy penalty and target area mask L 1 And (4) regularizing.
In one embodiment, the preset loss function is
L＝L m +α·L S +β·L 1
L j ＝T j max(0,m + -||v j ||) 2 +
λ(1-T j )max(0,||v j ||-m - ) 2
Wherein m is + =0.9 represents the lower limit of true positive, m - =0.1 for true negative upper bound, λ =0.5 for scaling factor, L m Represents the loss of separation of the capsule network, L S Represents significance detection two-class cross entropy loss, L 1 1-norm regularization, L, representing a mask of a target area j Denotes the loss of class j, T j Representing an indicator function, T j Only if 1 is the jth class, the hyper-parameter α is set to 1e-5 and β is set to 5e-6 to control the mask sparsity.
In one embodiment, the preprocessing module 602 is further configured to preprocess the radar target two-dimensional image to obtain an SAR grayscale image, and includes:
performing linear transformation and gray level enhancement on the radar target two-dimensional image to obtain an SAR gray level image;
the calculation process of the linear transformation is
Wherein x represents the original radar target two-dimensional image pixel input value, y represents the output image pixel value, and x max Representing the maximum value of the pixels, x, of the two-dimensional image of the original radar target min Representing the minimum value of the pixels of the two-dimensional image of the original radar target;
the calculation process of the gray enhancement is
Wherein I represents the gray value of the radar target two-dimensional image, O represents the gray value of the SAR gray image, and I min_count Gray value, I, representing radar target two-dimensional image with minimum occurrence max_count And representing the gray value with the largest occurrence number of the two-dimensional image of the radar target.
In one embodiment, the random enhancing and normalizing module 604 is further configured to perform random data enhancing and normalizing on the training set to obtain a normalized training set, including:
carrying out random data enhancement and normalization processing on the training set according to image rotation, gaussian white noise disturbance, random uniform noise replacement and a preset random data enhancement method implementation rate to obtain a normalized training set; the implementation rate of each method in random data enhancement is respectively 0.3.
For specific limitations of the radar target SAR image joint detection and identification device based on significance migration, reference may be made to the above limitations of the radar target SAR image joint detection and identification method based on significance migration, and details are not repeated here. All or part of the modules in the radar target SAR image joint detection and identification device based on the significance migration can be realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as shown in fig. 7. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to realize a radar target SAR image joint detection and identification method based on significance migration. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.
Those skilled in the art will appreciate that the architecture shown in fig. 7 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In an embodiment, a computer device is provided, comprising a memory storing a computer program and a processor implementing the steps of the method in the above embodiments when the processor executes the computer program.
In an embodiment, a computer storage medium is provided, on which a computer program is stored, which computer program, when being executed by a processor, is adapted to carry out the steps of the method of the above embodiments.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above may be implemented by hardware instructions of a computer program, which may be stored in a non-volatile computer-readable storage medium, and when executed, may include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), rambus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent application shall be subject to the appended claims.
Claims (10)
1. A radar target SAR image joint detection and identification method based on significance migration is characterized by comprising the following steps:
acquiring a two-dimensional image of a radar target to be identified;
preprocessing the radar target two-dimensional image to obtain an SAR gray level image;
dividing the SAR gray level image into a training set and a test set, and performing random data enhancement and normalization processing on the training set to obtain a normalized training set;
constructing a deep learning model; the deep learning model comprises a double-layer convolution, a capsule network, significance detection and a feature layer mask; the deep learning model is of a U-shaped structure;
training the deep learning model by using the normalized training set to obtain a trained deep learning model;
and identifying the test set according to the trained deep learning model to obtain an identification result.
2. The method of claim 1, wherein training the deep learning model using the normalized training set to obtain a trained deep learning model comprises:
obtaining a plurality of different training set samples from the normalized training set;
extracting target information of the different training set samples by using the double-layer convolution to obtain multi-scale target characteristics;
performing feature aggregation on the multi-scale target features according to the U-shaped structure of the deep learning model to obtain enhanced multi-scale target features;
migrating target importance information from an optical pre-training model by using the significance map as a pseudo label of the significance detection to obtain a target key area;
performing target and background separation on the enhanced multi-scale target features according to the feature layer mask to obtain target features;
inputting the target characteristics into the capsule network for characteristic conversion to obtain target space characteristics;
and training the deep learning model by using the target key area, the target spatial characteristics and a preset loss function to obtain the trained deep learning model.
3. The method of claim 1, wherein obtaining a saliency map comprises:
training the VGG16 model pre-trained on ImageNet under the MSTAR standard working condition data set to obtain a trained VGG16 model;
and extracting various significance maps from the trained VGG16 model according to the Guided Grad-Cam.
4. The method according to any of claims 1 to 3, wherein the pre-set loss function comprises interval loss, significance detection two-class cross-entropy loss and target area mask L of the capsule network 1 And (4) regularizing.
5. The method of claim 1, wherein the preset loss function is
L＝L m +α·L S +β·L 1
L j ＝T j max(0,m + -||v j ||) 2 +
λ(1-T j )max(0,||v j ||-m - ) 2
Wherein m is + =0.9 represents true positiveBoundary, m - =0.1 for true negative upper bound, λ =0.5 for scaling factor, L m Represents the spacing loss, L, of the capsule network S Represents significance detection two-class cross entropy loss, L 1 1-norm regularization, L, representing a mask of a target area j Denotes the loss of class j, T j Representing an indicator function, T j Only if 1 is the jth class, the hyper-parameter α is set to 1e-5 and β is set to 5e-6 to control the mask sparsity.
6. The method of claim 1, wherein preprocessing the radar target two-dimensional image to obtain a SAR grayscale image comprises:
performing linear transformation and gray level enhancement on the radar target two-dimensional image to obtain an SAR gray level image; the linear transformation is calculated by
Wherein x represents the original radar target two-dimensional image pixel input value, y represents the output image pixel value, and x max Representing the maximum value of the pixels, x, of the two-dimensional image of the original radar target min Representing the minimum value of the pixels of the two-dimensional image of the original radar target;
the gray scale enhancement is calculated by
Wherein I represents the gray value of the radar target two-dimensional image, O represents the gray value of the SAR gray image, I min_count Gray value, I, representing the radar target two-dimensional image occurring the least number of times max_count And representing the gray value with the largest occurrence number of the two-dimensional image of the radar target.
7. The method of claim 1, wherein performing stochastic data enhancement and normalization on the training set to obtain a normalized training set comprises:
carrying out random data enhancement and normalization processing on the training set according to image rotation, gaussian white noise disturbance, random uniform noise replacement and a preset random data enhancement method implementation rate to obtain a normalized training set; the implementation rate of each method in the random data enhancement is respectively 0.3.
8. A radar target SAR image joint detection and identification device based on significance migration is characterized by comprising:
the preprocessing module is used for acquiring a two-dimensional image of a radar target to be identified; preprocessing the radar target two-dimensional image to obtain an SAR gray level image;
the random enhancement and normalization module is used for dividing the SAR gray level image into a training set and a test set, and performing random data enhancement and normalization processing on the training set to obtain a normalized training set;
the model building and training module is used for building a deep learning model; the deep learning model comprises a double-layer convolution, a capsule network, significance detection and a feature layer mask; training the deep learning model by using the normalized training set to obtain a trained deep learning model;
and the recognition module is used for recognizing the test set according to the trained deep learning model to obtain a recognition result.
9. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor implements the steps of the method of any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210827227.8A CN115205680A (en) | 2022-07-14 | 2022-07-14 | Radar target SAR image joint detection and identification method based on significance migration |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210827227.8A CN115205680A (en) | 2022-07-14 | 2022-07-14 | Radar target SAR image joint detection and identification method based on significance migration |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115205680A true CN115205680A (en) | 2022-10-18 |
Family
ID=83580215
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210827227.8A Pending CN115205680A (en) | 2022-07-14 | 2022-07-14 | Radar target SAR image joint detection and identification method based on significance migration |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115205680A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116244662A (en) * | 2023-02-24 | 2023-06-09 | 中山大学 | Multisource elevation data fusion method, multisource elevation data fusion device, computer equipment and medium |
-
2022
- 2022-07-14 CN CN202210827227.8A patent/CN115205680A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116244662A (en) * | 2023-02-24 | 2023-06-09 | 中山大学 | Multisource elevation data fusion method, multisource elevation data fusion device, computer equipment and medium |
| CN116244662B (en) * | 2023-02-24 | 2023-11-03 | 中山大学 | Multisource elevation data fusion method, multisource elevation data fusion device, computer equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111860670B (en) | Domain adaptive model training method, image detection method, device, equipment and medium | |
| Fu et al. | An anchor-free method based on feature balancing and refinement network for multiscale ship detection in SAR images | |
| CN109801293B (en) | Remote sensing image segmentation method and device, storage medium and server | |
| Nieto-Hidalgo et al. | Two-stage convolutional neural network for ship and spill detection using SLAR images | |
| US10210418B2 (en) | Object detection system and object detection method | |
| CN110796048B (en) | Ship target real-time detection method based on deep neural network | |
| CN111950329A (en) | Target detection and model training method and device, computer equipment and storage medium | |
| CN109712128B (en) | Feature point detection method, feature point detection device, computer equipment and storage medium | |
| CN111080628A (en) | Image tampering detection method and device, computer equipment and storage medium | |
| CN102867196A (en) | Method for detecting complex sea-surface remote sensing image ships based on Gist characteristic study | |
| CN113536963B (en) | SAR image airplane target detection method based on lightweight YOLO network | |
| CN113989662A (en) | Remote sensing image fine-grained target identification method based on self-supervision mechanism | |
| Yue et al. | Texture extraction for object-oriented classification of high spatial resolution remotely sensed images using a semivariogram | |
| CN110163294B (en) | Remote sensing image change region detection method based on dimension reduction operation and convolution network | |
| CN114445356A (en) | Multi-resolution-based full-field pathological section image tumor rapid positioning method | |
| CN115205680A (en) | Radar target SAR image joint detection and identification method based on significance migration | |
| CN114897764A (en) | Pulmonary nodule false positive elimination method and device based on standardized channel attention | |
| CN112990107B (en) | Hyperspectral remote sensing image underwater target detection method and device and computer equipment | |
| CN113703048A (en) | Method and system for detecting high-resolution earthquake fault of antagonistic neural network | |
| CN111666813B (en) | Subcutaneous sweat gland extraction method of three-dimensional convolutional neural network based on non-local information | |
| KR101821770B1 (en) | Techniques for feature extraction | |
| CN112348750A (en) | SAR image change detection method based on threshold fusion and neighborhood voting | |
| CN116758419A (en) | Multi-scale target detection method, device and equipment for remote sensing image | |
| CN116416136A (en) | Data amplification method for ship target detection of visible light remote sensing image and electronic equipment | |
| CN111462059B (en) | Parallel processing method and device for intelligent target detection of fetal ultrasonic image |
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 |