CN111553421A - SAR equipment task failure cause reasoning method based on double-layer nested structure - Google Patents

SAR equipment task failure cause reasoning method based on double-layer nested structure Download PDF

Info

Publication number
CN111553421A
CN111553421A CN202010351812.6A CN202010351812A CN111553421A CN 111553421 A CN111553421 A CN 111553421A CN 202010351812 A CN202010351812 A CN 202010351812A CN 111553421 A CN111553421 A CN 111553421A
Authority
CN
China
Prior art keywords
abnormal
sar
feature
data 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.)
Granted
Application number
CN202010351812.6A
Other languages
Chinese (zh)
Other versions
CN111553421B (en
Inventor
凡时财
史顺周
邹见效
徐红兵
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202010351812.6A priority Critical patent/CN111553421B/en
Publication of CN111553421A publication Critical patent/CN111553421A/en
Application granted granted Critical
Publication of CN111553421B publication Critical patent/CN111553421B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/24323—Tree-organised classifiers
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/40—Analysis of texture
    • G06T7/41—Analysis of texture based on statistical description of texture
    • G06T7/45—Analysis of texture based on statistical description of texture using co-occurrence matrix computation
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30168—Image quality inspection

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Quality & Reliability (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention discloses an SAR equipment task failure cause reasoning method based on a double-layer nested structure, which relates to the technical field of SAR equipment guarantee and comprises the following steps: collecting SAR image data of known abnormal types; reconstructing sample categories, and extracting a sample set; calculating image quality evaluation characteristics; image feature transformation, normalization processing and class marking combination; integrated acquisition of training data sets
Figure DDA0002472178940000011
And
Figure DDA0002472178940000012
training a double-layer model; processing an unknown abnormal SAR image into a data set to be detected; and reasoning causes of the failure of the SAR equipment. The invention adopts double-layer stochastic SensenThe forest model combines failure causes which are easy to be wrongly classified, so that the total number of categories is reduced, the samples which are easy to be wrongly classified are secondarily classified by using the image local features on the basis of the result of the first-layer classifier, the precision of the random forest model for classifying abnormal images is enhanced, the problem that the effect of training the model by utilizing SAR image data of different terrains is poor is solved, and the accuracy of SAR equipment task failure cause reasoning is effectively improved.

Description

SAR equipment task failure cause reasoning method based on double-layer nested structure
Technical Field
The invention relates to the technical field of SAR equipment guarantee, in particular to an SAR equipment task failure cause reasoning method based on a double-layer nested structure.
Background
At the present time, the army is in the revolution from mechanized army to information army, and the information collecting, transmitting and processing capability is increasingly improved. SAR as a novel efficient information acquisition weapon has become a new approach for military observation and reconnaissance. SAR imaging is susceptible to various factors, and when an SAR radar executes a task and cannot obtain an image which is high in interpretation degree, good in effect and clear enough, namely when the task fails, fault detection and maintenance need to be carried out on the radar.
The traditional fault diagnosis and elimination technology mainly depends on expert experience, and has the problems of poor reliability, low accuracy and high false alarm rate, so that the completion of daily drilling tasks is influenced, even irrecoverable loss is caused to actual military operations, the problems seriously influence the logistics capacity and operational performance of radar equipment, and a failure cause reasoning method facing the task completion degree is urgently needed at present.
Disclosure of Invention
The invention aims to provide a SAR equipment task failure cause reasoning method based on a double-layer nested structure, which can alleviate the problems.
In order to alleviate the above problems, the technical scheme adopted by the invention is as follows:
a SAR equipment task failure cause reasoning method based on a double-layer nested structure comprises the following steps:
s1, collecting an SAR image data set of known abnormal types of K-type terrains, wherein the SAR image data set comprises a normal SAR image and an abnormal SAR image set, the abnormal SAR image set comprises P-type abnormal SAR images, the number of the abnormal SAR images is M/P, and M is the total number of the abnormal SAR images in the abnormal SAR image set;
s2, combining R type abnormal images I and R type abnormal images II in the abnormal SAR image group of each type of terrain, forming a sample set A1 together with other types of abnormal SAR images in the group, independently extracting the R type abnormal images I and the R type abnormal images II and forming a sample set A2, wherein the R type abnormal images I and the R type abnormal images II are in one-to-one correspondence and are all abnormal SAR images included in the abnormal SAR image group;
s3, constructing a feature data set D1 with N image evaluation indexes for each type of terrain sample set A1kThe feature set Z1 used for feature transformation and having N image evaluation indexes is used for constructing a feature data set D2 having T texture features for a sample set A2 of the feature data setkThe feature set Z2 is used for feature transformation and has T texture features, the N image evaluation indexes comprise an SAR image evaluation index based on a surface target, an SAR image texture feature index based on a gray level co-occurrence matrix and a gradient image evaluation index, and the T texture features are texture features based on a gray level-gradient co-occurrence matrix;
s4, for each type of terrain, feature data set D1kCarrying out feature transformation, normalization processing and class label merging relative to the feature set Z1 to obtain a data set I, and carrying out feature data set D2 on the data set IkCarrying out feature transformation, normalization processing and class mark merging relative to the feature set Z2 to obtain a data set II;
s5, integrating the data sets I of various terrains to obtain a training data set
Figure BDA0002472178920000021
S6, integrating the data sets II of various terrains to obtain a training data set
Figure BDA0002472178920000022
S7, training the double-layer model according to the training data set
Figure BDA0002472178920000023
Training by using a random forest algorithm to obtain a first-layer classifier of the double-layer model, and obtaining a first-layer classifier of the double-layer model according to a training data set
Figure BDA0002472178920000024
Training by using a random forest algorithm to obtain a second-layer classifier of the double-layer model;
s8, acquiring a plurality of unknown abnormal SAR images of the K-type terrain, and preprocessing the plurality of unknown abnormal SAR images of each type of terrain to obtain a feature matrix with the N image evaluation indexes
Figure BDA0002472178920000025
And having said T texture features
Figure BDA0002472178920000026
Feature matrix of K-type terrain
Figure BDA0002472178920000027
Obtaining a data set D to be measured after integrationde3Feature matrix of K-type terrain
Figure BDA0002472178920000028
Obtaining a data set D to be measured after integrationde4；
S9, collecting the data set D to be measuredde3Inputting a two-layer model, using its first-layer classifier to the data set D to be measuredde3Classifying to obtain a first classification result, if the abnormal type in the R-type abnormal image in the step S2 does not exist in the first classification result, taking the first classification result as a reasoning result of SAR equipment task failure cause, and finishing reasoning, otherwise, continuing to execute the step S10;
s10, the first classification result is compared with the data set D to be measuredde4Corresponding to the same data, extracting and forming a characteristic data set Ar, taking the residual data in the first classification result as the reasoning result of the first part of SAR equipment task failure cause, inputting the characteristic data set Ar into a second layer classifier of the double-layer model to obtain a second classification result, and taking the second classification result as the second part of SAR equipment task failure cause reasoning resultAnd (5) reasoning a result of the equipment task failure cause, and finishing reasoning.
The technical effect of the technical scheme is as follows: the method adopts a double-layer random forest model to combine failure reasons which are easy to be wrongly classified, so that the total number of categories is reduced, and the samples which are easy to be wrongly classified are secondarily classified by using image local characteristics on the basis of the result of a first-layer classifier, so that the precision of the random forest model for classifying abnormal images is enhanced, the problem of poor effect of training the model by utilizing SAR image data of different terrains is solved, and the accuracy of SAR equipment task failure cause reasoning is effectively improved; and for special conditions of abnormal categories which are easy to be wrongly classified, the data are separately classified again by using the characteristics which can better reflect local information, so that the accurate judgment of failure causes under the task failure scene of the airborne SAR is realized.
Further, in the step S1, the abnormal SAR image includes N1 fault types, N2 interference types, and N1 × N2 fault interference superposition types.
The technical effect of the technical scheme is as follows: SAR radar imaging is susceptible to various factors, the conditions of low image interpretation degree, poor effect and poor definition often occur, and the influence on the actual work of the SAR radar can be comprehensively considered by dividing the influence factors.
Further, in step S2, it is determined whether the abnormal image i is similar to the abnormal image ii according to the abnormal type flag of the image, which is a flag made when the image is collected.
The technical effect of the technical scheme is as follows: the images which are easy to be wrongly distinguished are convenient to distinguish, and the training of the subsequent model is convenient.
Further, the feature data set D1kThe construction method of the sum feature set Z1 comprises the following steps: for each abnormal SAR image in the sample set A1, N image evaluation indexes are calculated, and a feature data set D1 is constructed according to the abnormal type labels and the N image evaluation indexes of the abnormal SAR images in the sample set A1kComputing a feature data set D1kN image evaluation indexes of normal SAR image of the terrain and form a feature setZ1；
Feature data set D2kThe construction method of the sum feature set Z2 comprises the following steps: for each abnormal SAR image in the sample set A2, calculating T texture features based on a gray-gradient co-occurrence matrix, and constructing a feature data set D2 according to the abnormal type marks of the abnormal SAR images in the sample set A2 and the T texture features based on the gray-gradient co-occurrence matrixkComputing a feature data set D2kThe T types of normal SAR images of the terrain are based on texture features of a gray-gradient co-occurrence matrix and form a feature set Z2.
The technical effect of the technical scheme is as follows: the SAR image evaluation index based on the surface target, the SAR image texture characteristic index based on the gray level co-occurrence matrix and the gradient image evaluation index are objective image evaluation indexes, and the image quality evaluation result can be guaranteed to have good stability.
Still further, the surface target-based SAR image evaluation indicators include a mean, a variance, a radiation resolution, an equivalent vision, and a sharpness, the gray level co-occurrence matrix-based SAR image texture feature indicators include an angular second moment, a contrast, a correlation, an entropy, an inverse difference, a sum mean, a sum variance, a sum entropy, a variation difference, a difference entropy, a mutual information metric, a maximum correlation coefficient, a maximum probability, a dissimilarity, a contrast, a median, a dark cluster, and a salient cluster, and the gradient image evaluation indicators include an energy gradient, an average gradient, a gray level difference product, a Brenner gradient, and a Laplacian gradient.
The technical effect of the technical scheme is as follows: the SAR image can be comprehensively evaluated by comprehensively considering the information of the surface target, texture, gradient and the like of the SAR image.
Still further, the texture features based on the gray-gradient co-occurrence matrix include small gradient dominance, large gradient dominance, gray distribution heterogeneity, gradient distribution heterogeneity, energy, gray mean, gradient mean, gray standard deviation, gradient standard deviation, correlation, gray entropy, gradient entropy, mixture entropy, difference moment, inverse difference moment.
The technical effect of the technical scheme is as follows: the texture features based on the gray-gradient co-occurrence matrix can better reflect local information of the SAR image, so that overall statistical information and local difference information of the SAR image are considered, and multi-level and multi-index image quality evaluation indexes are established to comprehensively infer the task failure cause by combining the overall characteristic evaluation index and the local characteristic evaluation index of the SAR image, thereby comprehensively evaluating the overall quality and the local distortion condition of the SAR image.
Further, in step S8, the preprocessing the unknown abnormal SAR image includes:
constructing a feature data set D3, comprising: for each unknown abnormal SAR image, calculating the N image evaluation indexes, and constructing a feature data set D3 according to the N image evaluation indexes of each abnormal SAR image;
constructing a feature data set D4, comprising: for each unknown abnormal SAR image, calculating the T texture features of the unknown abnormal SAR image, and constructing a feature data set D4 according to the T texture features of each abnormal SAR image;
performing feature transformation on the feature data set D3 relative to the feature set Z1, and performing normalization processing on the transformation result to obtain the feature matrix
Figure BDA0002472178920000041
Performing feature transformation on the feature data set D4 relative to the feature set Z2, and performing normalization processing on the transformation result to obtain the feature matrix
Figure BDA0002472178920000042
The technical effect of the technical scheme is as follows: the method solves the problem that the accuracy of the inference result is influenced due to inconsistent performance of the characteristic factors of the inference problem of SAR task failure causes under different terrains by a new characteristic transformation strategy of mapping the original characteristic space to a difference space with the standard normal image characteristics, and can accelerate the training of a subsequent model through data normalization.
Further, the feature transformation is specifically: and performing difference operation on the features in the feature data set and the features in the feature set for feature transformation in a one-to-one correspondence manner, and taking the obtained difference result as the transformation result of the feature data set.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a flow chart of a method of a two-level nested structure model employed in the present invention;
FIG. 2 is a flowchart of a specific embodiment of the SAR equipment task failure cause reasoning method based on a double-layer nested structure;
FIG. 3 is an example of a SAR image used by the present invention for training and testing.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, 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 some, but not all, embodiments of the present invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. 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.
Examples
Referring to fig. 1 and fig. 2, the present embodiment provides a method for reasoning cause of task failure of SAR equipment based on a double-layer nested structure, including the following steps:
s1, collecting an SAR image data set of known abnormal types of K types of terrains, wherein the SAR image data set comprises a normal SAR image and an abnormal SAR image set, the abnormal SAR image set comprises P types of abnormal SAR images, the number of the abnormal SAR images is M/P, and M is the total number of the abnormal SAR images in the abnormal SAR image set.
In the present embodiment, the SAR image data of known abnormal type is from some type of onboard SAR radar, where K is 6, 6 types of terrain are mountainous area, typical building, lake, hill, island and small airport, respectively, and M is 290, 6 types of terrain total 1740 abnormal SAR images and 6 normal SAR images, as shown in fig. 3.
In the present embodiment, the total number of abnormal types P is 29, where the number of fault types N1 is 4, the number of interference types N2 is 5, the number of superimposed fault interference types N1 × N2 is 20, and the number of superimposed fault interference types is obtained by superimposing 5 types of interference with 4 types of faults. For each terrain, the number of abnormal images of each type is M/P-10, and the total number is 1740.
And S2, combining the R type abnormal images I and the R type abnormal images II in the abnormal SAR image group of each type of terrain, forming a sample set A1 together with other types of abnormal SAR images in the group, and separately extracting the R type abnormal images I and the R type abnormal images II to form a sample set A2, wherein the R type abnormal images I and the R type abnormal images II are in one-to-one correspondence and are all the abnormal SAR images included in the abnormal SAR image group.
In this embodiment, R is 3, which is the dominant wave error plus the deception interference, the platform vibration plus the deception interference, and the inertial navigation error plus the deception interference, respectively, these 3 anomalies are easily separated from the dominant wave error, the platform vibration, and the inertial navigation error, respectively, and these 3 categories are merged into 3 categories of the dominant wave error, the platform vibration, and the inertial navigation error, respectively, and then constitute a sample set a1 together with other categories of anomalous SAR images in the anomalous SAR image group. These 6 abnormal image structure sample sets a2 were extracted separately, where a2 has S-360 images.
In the present embodiment, it is determined whether the abnormal image i is similar to the abnormal image ii based on the abnormal type class flag of the image, which is a flag made at the time of collecting the image.
Whether the abnormal image I is similar to the abnormal image II or not is judged through manual judgment, for example, the deception jamming effect means that a false target exists in the image, which is not beneficial to military judgment, and when the deception jamming is weak, the two kinds of anomalies, namely a single fault condition and fault superposition weak deception jamming, are close to each other, and the difference is difficult to judge by human eyes.
S3, constructing a feature data set D1 with N image evaluation indexes for each type of terrain sample set A1kThe feature set Z1 used for feature transformation and having N image evaluation indexes is used for constructing a feature data set D2 having T texture features for a sample set A2 of the feature data setkThe feature set Z2 is used for feature transformation and has T texture features, the N image evaluation indexes comprise SAR image evaluation indexes based on a surface target, SAR image texture feature indexes based on a gray level co-occurrence matrix and gradient image evaluation indexes, and the T texture features are texture features based on the gray level-gradient co-occurrence matrix.
For the kth terrain, calculating N image evaluation indexes of each abnormal SAR image in a sample set A1, and constructing a feature data set D1 of M abnormal SAR images in the kth terrain according to the image evaluation indexes and the abnormal type class marksk；
Figure BDA0002472178920000061
Calculating N image evaluation indexes of the normal SAR image, and forming a feature set Z1;
Figure BDA0002472178920000071
wherein,
Figure BDA0002472178920000072
Xka feature data set (i.e. a feature matrix) representing an abnormal image of the a1 sample set,
Figure BDA0002472178920000073
a column vector formed by the characteristics corresponding to the n-th evaluation index of the M abnormal images, YkA column vector which represents the abnormal type constitution corresponding to the M abnormal images,
Figure BDA0002472178920000074
Figure BDA0002472178920000075
class labels that are exception types;
in this embodiment, the image quality evaluation characteristic is set to be N ═ 30, where the SAR image evaluation indexes based on the surface target include mean, variance, radiation resolution, equivalent view, and sharpness, the SAR image texture characteristic indexes based on the gray level co-occurrence matrix include angular second moment, contrast, correlation, entropy, inverse difference, sum average, sum variance, sum entropy, variation difference, difference entropy, mutual information metric, maximum correlation coefficient, maximum probability, difference, contrast, median, dark cluster, and salient cluster, and the conventional gradient image evaluation indexes include energy gradient, average gradient, gray level difference product, Brenner gradient, and Laplacian gradient.
S images are set in a sample set A2, T texture features based on a gray-gradient co-occurrence matrix of each abnormal SAR image in the sample set A2 are calculated, and a feature data set D2 is constructed by utilizing the texture features based on the gray-gradient co-occurrence matrix and the abnormal type marks of the imagesk；
Figure BDA0002472178920000076
The normal image of the kth terrain is characterized by:
Figure BDA0002472178920000077
wherein,
Figure BDA0002472178920000078
Figure BDA0002472178920000079
a feature data set (i.e. a feature matrix) representing an abnormal image of the a2 sample set,
Figure BDA00024721789200000710
a column vector formed by the characteristics corresponding to the t-th evaluation index of the S abnormal images,
Figure BDA00024721789200000711
a column vector which represents the abnormal type constitution corresponding to the S abnormal images,
Figure BDA00024721789200000712
s＝1,2,…,S，
Figure BDA00024721789200000713
class labels that are exception types;
in this embodiment, the image quality assessment feature is set to T15, and is mainly based on texture features of the gray-gradient co-occurrence matrix, including small gradient dominance, large gradient dominance, gray distribution heterogeneity, gradient distribution heterogeneity, energy, gray mean, gradient mean, gray standard difference, gradient standard difference, correlation, gray entropy, gradient entropy, mixed entropy, differential moment, and inverse differential moment.
S4, for each type of terrain, feature data set D1kCarrying out feature transformation, normalization processing and class label merging relative to the feature set Z1 to obtain a data set I, and carrying out feature data set D2 on the data set IkAnd (5) performing feature transformation, normalization processing and class mark merging relative to the feature set Z2 to obtain a data set II.
In this embodiment, the feature transformation refers to performing difference operation on each feature in the feature data set and each feature in the feature set for feature transformation in a one-to-one correspondence manner, and taking an obtained difference result as a transformation result of the feature data set, which is specifically as follows:
for the feature data set D1kIs characterized by that it has
Figure BDA0002472178920000081
For the feature data set D2kIs characterized by that it has
Figure BDA0002472178920000082
Wherein, X'kIs XkThe feature matrix after the feature transformation is carried out,
Figure BDA0002472178920000091
is composed of
Figure BDA0002472178920000092
And (5) feature matrix after feature transformation.
Then respectively aligning the feature matrices X′kAnd
Figure BDA0002472178920000093
normalizing by columns to obtain normalized feature matrix X'kAnd
Figure BDA0002472178920000094
Figure BDA0002472178920000095
Figure BDA0002472178920000096
wherein, the normalization operation according to the columns is as follows:
Figure BDA0002472178920000097
Figure BDA0002472178920000098
therein, maxnFor the n-th dimension, i.e. feature matrix X′kMaximum of the n-th element, minnFor the n-th dimension, i.e. feature matrix X′kMinimum value of the nth column element;
Figure BDA0002472178920000099
for the t-th dimension, i.e. feature matrix
Figure BDA00024721789200000910
The maximum value of the element in the t-th column,
Figure BDA00024721789200000911
for the t-th dimension, i.e. feature matrix
Figure BDA00024721789200000912
The t-th column element minimum;
finally, the normalized feature matrix X "kAnd class label column vector YkMerging to obtain the characteristic data set D of N abnormal images under the kth terrain "kI.e. data set i; will be provided with
Figure BDA00024721789200000913
And class label column vector
Figure BDA00024721789200000914
Merging to obtain the characteristic data set of S abnormal images in the kth terrain
Figure BDA00024721789200000915
Namely a data set II;
data set I
Figure BDA00024721789200000916
Data set II
Figure BDA00024721789200000917
S5、Data sets of various terrains ID'kIntegrating to obtain a training data set
Figure BDA0002472178920000101
Wherein G ═ M × K.
S6, collecting data II of various terrains
Figure BDA0002472178920000102
Integrating to obtain a training data set
Figure BDA0002472178920000103
Wherein F ═ sxk.
S7, training the double-layer model according to the training data set
Figure BDA0002472178920000104
Training by using a random forest algorithm to obtain a first-layer classifier of the double-layer model, and obtaining a first-layer classifier of the double-layer model according to a training data set
Figure BDA0002472178920000105
And training by using a random forest algorithm to obtain a second-layer classifier of the double-layer model.
S8, acquiring a plurality of unknown abnormal SAR images of the K-type terrain, and preprocessing the plurality of unknown abnormal SAR images of each type of terrain to obtain a feature matrix with N image evaluation indexes
Figure BDA0002472178920000106
And having T texture features
Figure BDA0002472178920000107
Feature matrix of K-type terrain
Figure BDA0002472178920000108
Obtaining a data set D to be measured after integrationde3Feature matrix of K-type terrain
Figure BDA0002472178920000109
Obtaining a data set D to be measured after integrationde4；
In this embodiment, the process of preprocessing the unknown abnormal SAR image includes the same image processing process as that in the model training process, including:
constructing a feature data set D3, comprising: for each unknown abnormal SAR image, calculating the N image evaluation indexes, and constructing a feature data set D3 according to the N image evaluation indexes of each abnormal SAR image;
constructing a feature data set D4, comprising: for each unknown abnormal SAR image, calculating the T texture features of the unknown abnormal SAR image, and constructing a feature data set D4 according to the T texture features of each abnormal SAR image;
performing feature transformation on the feature data set D3 relative to the feature set Z1, and performing normalization processing on the transformation result to obtain the feature matrix
Figure BDA00024721789200001010
Performing feature transformation on the feature data set D4 relative to the feature set Z2, and performing normalization processing on the transformation result to obtain the feature matrix
Figure BDA00024721789200001011
Then, the feature matrix of the K-type terrain
Figure BDA00024721789200001012
Obtaining a data set D to be measured after integrationde3Feature matrix of K-type terrain
Figure BDA0002472178920000111
Obtaining a data set D to be measured after integrationde4：
Figure BDA0002472178920000112
Figure BDA0002472178920000113
In this example, 500 abnormal SAR images are selected as Q abnormal SAR images of unknown abnormal types for verification.
S9, collecting the data set D to be measuredde3Inputting a two-layer model, using its first-layer classifier to the data set D to be measuredde3And (4) classifying to obtain a first classification result, namely an abnormal type corresponding to each of the Q abnormal SAR images, if the abnormal type in the R abnormal images in the step (S2) does not exist in the first classification result (namely the abnormal type is not the R type which is easy to be wrongly classified), the abnormal type is used as an inference result of the SAR equipment task failure cause, the inference is finished, and otherwise, the step (S10) is continuously executed.
S10, the first classification result is compared with the data set D to be measuredde4And taking out the same corresponding data and forming a characteristic data set Ar, taking the residual data in the first classification result as a first part SAR equipment task failure cause reasoning result, inputting the characteristic data set Ar into a second layer classifier of the double-layer model to obtain a second classification result, taking the second classification result as a second part SAR equipment task failure cause reasoning result, and finishing reasoning.
The final result can be expressed as Yde＝{yde_1,yde_2,…,yde_QAnd each element represents the abnormal type of the corresponding numbered abnormal SAR image.
In the present embodiment, the generalization ability of the two-layer model is evaluated using the Accuracy (ACC).
Figure BDA0002472178920000114
Wherein TN, TP, FN and FP represent the number of true negative, true positive, false negative and false positive, respectively.
Table 1 shows the accuracy index result of the inference method for the cause of task failure of the SAR equipment based on the double-layer nested structure in this embodiment.
TABLE 1
Number of training samples Number of samples tested Test set ACC
Adaboost 1740 500 0.149
Decision tree 1740 500 0.300
Support vector machine 1740 500 0.497
K nearest neighbor 1740 500 0.586
Random forest 1740 500 0.866
Double-layer nested random forest 1740 500 0.930
The accuracy comparison of the algorithm models shows that the constructed multi-classification random forest model based on the double-layer nested structure has the highest classification precision, and the purpose of the invention is realized.
The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes will occur to those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (8)

1. A SAR equipment task failure cause reasoning method based on a double-layer nested structure is characterized by comprising the following steps:
s1, collecting an SAR image data set of known abnormal types of K-type terrains, wherein the SAR image data set comprises a normal SAR image and an abnormal SAR image set, the abnormal SAR image set comprises P-type abnormal SAR images, the number of the abnormal SAR images is M/P, and M is the total number of the abnormal SAR images in the abnormal SAR image set;
s2, combining R type abnormal images I and R type abnormal images II in the abnormal SAR image group of each type of terrain, forming a sample set A1 together with other types of abnormal SAR images in the group, independently extracting the R type abnormal images I and the R type abnormal images II and forming a sample set A2, wherein the R type abnormal images I and the R type abnormal images II are in one-to-one correspondence and are all abnormal SAR images included in the abnormal SAR image group;
s3, constructing a feature data set D1 with N image evaluation indexes for each type of terrain sample set A1kThe feature set Z1 used for feature transformation and having N image evaluation indexes is used for constructing a sample set A2 with T patternsFeature data set D2 of physical featureskThe feature set Z2 is used for feature transformation and has T texture features, the N image evaluation indexes comprise an SAR image evaluation index based on a surface target, an SAR image texture feature index based on a gray level co-occurrence matrix and a gradient image evaluation index, and the T texture features are texture features based on a gray level-gradient co-occurrence matrix;
s4, for each type of terrain, feature data set D1kCarrying out feature transformation, normalization processing and class label merging relative to the feature set Z1 to obtain a data set I, and carrying out feature data set D2 on the data set IkCarrying out feature transformation, normalization processing and class mark merging relative to the feature set Z2 to obtain a data set II;
s5, integrating the data sets I of various terrains to obtain a training data set
Figure FDA0002472178910000011
S6, integrating the data sets II of various terrains to obtain a training data set
Figure FDA0002472178910000012
S7, training the double-layer model according to the training data set
Figure FDA0002472178910000013
Training by using a random forest algorithm to obtain a first-layer classifier of the double-layer model, and obtaining a first-layer classifier of the double-layer model according to a training data set
Figure FDA0002472178910000014
Training by using a random forest algorithm to obtain a second-layer classifier of the double-layer model;
s8, acquiring a plurality of unknown abnormal SAR images of the K-type terrain, and preprocessing the plurality of unknown abnormal SAR images of each type of terrain to obtain a feature matrix with the N image evaluation indexes
Figure FDA0002472178910000015
And having said T texture features
Figure FDA0002472178910000021
Feature matrix of K-type terrain
Figure FDA0002472178910000022
Obtaining a data set D to be measured after integrationde3Feature matrix of K-type terrain
Figure FDA0002472178910000023
Obtaining a data set D to be measured after integrationde4；
S9, collecting the data set D to be measuredde3Inputting a two-layer model, using its first-layer classifier to the data set D to be measuredde3Classifying to obtain a first classification result, if the abnormal type in the R-type abnormal image in the step S2 does not exist in the first classification result, taking the first classification result as a reasoning result of SAR equipment task failure cause, and finishing reasoning, otherwise, continuing to execute the step S10;
s10, the first classification result is compared with the data set D to be measuredde4And taking out the same corresponding data and forming a characteristic data set Ar, taking the residual data in the first classification result as a first part SAR equipment task failure cause reasoning result, inputting the characteristic data set Ar into a second layer classifier of the double-layer model to obtain a second classification result, taking the second classification result as a second part SAR equipment task failure cause reasoning result, and finishing reasoning.
2. The SAR equipment task failure cause reasoning method based on the double-layer nested structure as claimed in claim 1, wherein in the step S1, the abnormal SAR image includes N1 fault types, N2 interference types and N1 XN 2 fault interference superposition types.
3. The SAR equipment task failure cause reasoning method based on the double-layer nested structure as claimed in claim 1, wherein in step S2, it is determined whether an abnormal image I is similar to an abnormal image II according to an abnormal type mark of the image, where the abnormal type mark is a mark made when the image is collected.
4. The SAR equipment task failure cause reasoning method based on the double-layer nested structure as claimed in claim 1,
feature data set D1kThe construction method of the sum feature set Z1 comprises the following steps: for each abnormal SAR image in the sample set A1, N image evaluation indexes are calculated, and a feature data set D1 is constructed according to the abnormal type labels and the N image evaluation indexes of the abnormal SAR images in the sample set A1kComputing a feature data set D1kN image evaluation indexes of the normal SAR image of the terrain to which the image belongs form a feature set Z1;
feature data set D2kThe construction method of the sum feature set Z2 comprises the following steps: for each abnormal SAR image in the sample set A2, calculating T texture features based on a gray-gradient co-occurrence matrix, and constructing a feature data set D2 according to the abnormal type marks of the abnormal SAR images in the sample set A2 and the T texture features based on the gray-gradient co-occurrence matrixkComputing a feature data set D2kThe T types of normal SAR images of the terrain are based on texture features of a gray-gradient co-occurrence matrix and form a feature set Z2.
5. The SAR equipment task failure cause reasoning method based on the double-layer nested structure is characterized in that the SAR image evaluation indexes based on the plane target comprise a mean value, a variance, a radiation resolution, an equivalent vision and a sharpness, the SAR image texture feature indexes based on the gray level co-occurrence matrix comprise an angular second moment, a contrast, a correlation, an entropy, an inverse difference, a mean, a variance, a sum entropy, a variation difference, a difference entropy, a mutual information measure, a maximum correlation coefficient, a maximum probability, a difference, a contrast, a median, a dark cluster and a salient cluster, and the gradient image evaluation indexes comprise an energy gradient, a mean gradient, a gray level difference product, a Brenner gradient and a Laplacian gradient.
6. The SAR equipment task failure cause reasoning method based on the double-layer nested structure as claimed in claim 5, wherein the texture features based on the gray-gradient co-occurrence matrix include small gradient dominance, large gradient dominance, gray distribution heterogeneity, gradient distribution heterogeneity, energy, gray mean, gradient mean, gray standard deviation, gradient standard deviation, correlation, gray entropy, gradient entropy, mixed entropy, difference moment, and inverse difference moment.
7. The SAR equipment task failure cause reasoning method based on the double-layer nested structure as claimed in claim 6, wherein in the step S8, the process of preprocessing the unknown abnormal SAR image comprises:
constructing a feature data set D3, comprising: for each unknown abnormal SAR image, calculating the N image evaluation indexes, and constructing a feature data set D3 according to the N image evaluation indexes of each abnormal SAR image;
constructing a feature data set D4, comprising: for each unknown abnormal SAR image, calculating the T texture features of the unknown abnormal SAR image, and constructing a feature data set D4 according to the T texture features of each abnormal SAR image;
performing feature transformation on the feature data set D3 relative to the feature set Z1, and performing normalization processing on the transformation result to obtain the feature matrix
Figure FDA0002472178910000031
Performing feature transformation on the feature data set D4 relative to the feature set Z2, and performing normalization processing on the transformation result to obtain the feature matrix
Figure FDA0002472178910000032
8. The SAR equipment task failure cause reasoning method based on the double-layer nested structure as claimed in claim 7, wherein the feature transformation specifically comprises: and performing difference operation on the features in the feature data set and the features in the feature set for feature transformation in a one-to-one correspondence manner, and taking the obtained difference result as the transformation result of the feature data set.
CN202010351812.6A 2020-04-28 2020-04-28 SAR equipment task failure cause reasoning method based on double-layer nested structure Expired - Fee Related CN111553421B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010351812.6A CN111553421B (en) 2020-04-28 2020-04-28 SAR equipment task failure cause reasoning method based on double-layer nested structure

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010351812.6A CN111553421B (en) 2020-04-28 2020-04-28 SAR equipment task failure cause reasoning method based on double-layer nested structure

Publications (2)

Publication Number Publication Date
CN111553421A true CN111553421A (en) 2020-08-18
CN111553421B CN111553421B (en) 2022-03-08

Family

ID=72002626

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010351812.6A Expired - Fee Related CN111553421B (en) 2020-04-28 2020-04-28 SAR equipment task failure cause reasoning method based on double-layer nested structure

Country Status (1)

Country Link
CN (1) CN111553421B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116863492A (en) * 2023-09-04 2023-10-10 山东正禾大教育科技有限公司 Mobile digital publishing system
CN118427825A (en) * 2024-07-04 2024-08-02 北京天融信网络安全技术有限公司 Secure adaptive federal learning method, electronic device, and storage medium

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2784537A1 (en) * 2013-05-15 2014-10-01 Institute of Electronics, Chinese Academy of Sciences Inversion method and apparatus based on polarimetric interferometric synthetic aperture radar
CN104217217A (en) * 2014-09-02 2014-12-17 武汉睿智视讯科技有限公司 Vehicle logo detection method and system based on two-layer classification
CN104680141A (en) * 2015-02-13 2015-06-03 华中师范大学 Motion unit layering-based facial expression recognition method and system
CN105139395A (en) * 2015-08-19 2015-12-09 西安电子科技大学 SAR image segmentation method based on wavelet pooling convolutional neural networks
CN110569624A (en) * 2019-09-20 2019-12-13 哈尔滨工业大学 Forest three-layer scattering model determining and analyzing method suitable for PolInSAR inversion
CN110751201A (en) * 2019-10-16 2020-02-04 电子科技大学 SAR equipment task failure cause reasoning method based on textural feature transformation

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2784537A1 (en) * 2013-05-15 2014-10-01 Institute of Electronics, Chinese Academy of Sciences Inversion method and apparatus based on polarimetric interferometric synthetic aperture radar
CN104217217A (en) * 2014-09-02 2014-12-17 武汉睿智视讯科技有限公司 Vehicle logo detection method and system based on two-layer classification
CN104680141A (en) * 2015-02-13 2015-06-03 华中师范大学 Motion unit layering-based facial expression recognition method and system
CN105139395A (en) * 2015-08-19 2015-12-09 西安电子科技大学 SAR image segmentation method based on wavelet pooling convolutional neural networks
CN110569624A (en) * 2019-09-20 2019-12-13 哈尔滨工业大学 Forest three-layer scattering model determining and analyzing method suitable for PolInSAR inversion
CN110751201A (en) * 2019-10-16 2020-02-04 电子科技大学 SAR equipment task failure cause reasoning method based on textural feature transformation

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
M. NEUMANN 等: "Assessing Performance of L- and P-Band Polarimetric Interferometric SAR Data in Estimating Boreal Forest Above-Ground Biomass", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 *
史顺周: "基于改进深度森林的SAR任务失效成因推理方法研究", 《万方数据知识服务平台》 *
詹森 等: "基于纹理特征的高分辨率合成孔径雷达影像单体建筑物震害信息识别", 《科学技术与工程》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116863492A (en) * 2023-09-04 2023-10-10 山东正禾大教育科技有限公司 Mobile digital publishing system
CN116863492B (en) * 2023-09-04 2023-11-21 山东正禾大教育科技有限公司 Mobile digital publishing system
CN118427825A (en) * 2024-07-04 2024-08-02 北京天融信网络安全技术有限公司 Secure adaptive federal learning method, electronic device, and storage medium

Also Published As

Publication number Publication date
CN111553421B (en) 2022-03-08

Similar Documents

Publication Publication Date Title
CN107341488B (en) SAR image target detection and identification integrated method
Yun et al. PNN based crop disease recognition with leaf image features and meteorological data
CN101950364A (en) Remote sensing image change detection method based on neighbourhood similarity and threshold segmentation
Nurmasari et al. Oil palm plantation detection in Indonesia using Sentinel-2 and Landsat-8 optical satellite imagery (case study: Rokan Hulu regency, Riau Province)
CN106886760A (en) A kind of EO-1 hyperion Ship Detection combined based on empty spectrum information
CN116469020A (en) Unmanned aerial vehicle image target detection method based on multiscale and Gaussian Wasserstein distance
CN107341813A (en) SAR image segmentation method based on structure learning and sketch characteristic inference network
CN109191418A (en) A kind of method for detecting change of remote sensing image based on contraction self-encoding encoder feature learning
CN106897722A (en) A kind of trademark image retrieval method based on region shape feature
CN109146890A (en) The Anomaly target detection method of high spectrum image based on filter
Jiang et al. Focal-test-based spatial decision tree learning: A summary of results
Tamim et al. A simple and efficient approach for coarse segmentation of Moroccan coastal upwelling
CN111553421B (en) SAR equipment task failure cause reasoning method based on double-layer nested structure
CN116152678A (en) Marine disaster-bearing body identification method based on twin neural network under small sample condition
CN111860359A (en) Point cloud classification method based on improved random forest algorithm
CN115131580A (en) Space target small sample identification method based on attention mechanism
CN116958606A (en) Image matching method and related device
CN106951924B (en) Seismic coherence body image fault automatic identification method and system based on AdaBoost algorithm
CN112734695B (en) SAR image change detection method based on regional enhancement convolutional neural network
CN109948556A (en) Crops Classification recognition methods and system
CN110147749A (en) A kind of civil aviaton's missing baggage auxiliary method for retrieving
CN111428752B (en) Visibility detection method based on infrared image
LU103119B1 (en) Object-oriented method and system for classifying earth surface coverage through high-resolution remote sensing images
Corves et al. Mapping the reliability of satellite-derived landcover maps—an example from The Central Brazilian Amazon Basin
CN113674225A (en) Power equipment fault detection method based on convolutional neural network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20220308

CF01 Termination of patent right due to non-payment of annual fee