CN116401588A - Radiation source individual analysis method and device based on deep network - Google Patents
Radiation source individual analysis method and device based on deep network Download PDFInfo
- Publication number
- CN116401588A CN116401588A CN202310675856.8A CN202310675856A CN116401588A CN 116401588 A CN116401588 A CN 116401588A CN 202310675856 A CN202310675856 A CN 202310675856A CN 116401588 A CN116401588 A CN 116401588A
- Authority
- CN
- China
- Prior art keywords
- sample
- sample set
- radiation source
- source signal
- labeling area
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000005855 radiation Effects 0.000 title claims abstract description 172
- 238000004458 analytical method Methods 0.000 title claims abstract description 63
- 238000002372 labelling Methods 0.000 claims abstract description 108
- 238000006243 chemical reaction Methods 0.000 claims abstract description 73
- 238000004364 calculation method Methods 0.000 claims abstract description 43
- 238000000034 method Methods 0.000 claims abstract description 30
- 238000012545 processing Methods 0.000 claims description 39
- 230000002159 abnormal effect Effects 0.000 claims description 36
- 238000003062 neural network model Methods 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- 238000012300 Sequence Analysis Methods 0.000 claims description 6
- 230000000873 masking effect Effects 0.000 claims description 6
- 230000004044 response Effects 0.000 description 14
- 239000011159 matrix material Substances 0.000 description 10
- 230000008859 change Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000012732 spatial analysis Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000013136 deep learning model Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000007613 environmental effect Effects 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 229910000831 Steel Inorganic materials 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000013095 identification testing Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
- 238000012731 temporal analysis Methods 0.000 description 1
- 238000000700 time series analysis Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
- 238000007794 visualization technique Methods 0.000 description 1
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
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0475—Generative networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Investigating Or Analysing Biological Materials (AREA)
Abstract
The invention provides a radiation source individual analysis method and a radiation source individual analysis device based on a deep network, and relates to the field of radiation source individual analysis, wherein the method comprises the following steps: acquiring an original radiation source signal sample set; carrying out sample characteristic labeling on an original radiation source signal sample set through Grad-CAM calculation to obtain a high-reaction labeling area of each radiation source signal sample; analyzing the high-reaction labeling area of each radiation source signal sample to obtain a normal sample set and an unlabeled problem sample set; carrying out labeling description on the unlabeled problem sample set to obtain a labeled problem sample set; expanding the normal sample set and the labeling problem sample set to obtain an expanded sample set; and placing the expanded sample set into the original radiation source signal sample set to obtain a new radiation source signal sample set. The method can visually present the characteristic positions of the individual samples on one hand and can carry out interpretable expression on the radiation source individuals on the other hand.
Description
Technical Field
The invention relates to the field of radiation source individual analysis, in particular to a radiation source individual analysis method and device based on a deep network.
Background
The radiation source identification is classified into a classification identification of different types of radiation sources and an individual identification of the same type of radiation sources. For different types of radiation sources, the signal characteristics of the radiation sources are generally greatly different, and the radiation sources can be distinguished through parameter identification such as signal frequency, bandwidth, modulation mode and the like; the characteristics of different radiation source individuals of the same type are very similar, and are difficult to distinguish by the traditional characteristic extraction mode. The conventional radiation source individual identification method generally extracts transient or steady state high-order features of signals first, and then performs individual distinction based on the features. On one hand, the prior method lacks of carrying out visual presentation on the characteristic position of an individual sample, and on the other hand, the prior method has poor interpretability on the individual radiation source. Therefore, there is a need for a deep network-based method for analyzing radiation source individuals, which can visually present the characteristic positions of individual samples on one hand and can interpret the radiation source individuals on the other hand.
Disclosure of Invention
The present invention is directed to a method and apparatus for analyzing radiation source individuals based on deep network, which can solve the above problems. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
in a first aspect, the present application provides a deep network-based radiation source individual analysis method, the method comprising:
acquiring an original radiation source signal sample set, the original radiation source signal sample set comprising a plurality of radiation source signal samples;
carrying out sample characteristic labeling on the original radiation source signal sample set through Grad-CAM calculation to obtain a high-reaction labeling area of each radiation source signal sample;
analyzing the high-reaction labeling area of each radiation source signal sample to obtain a normal sample set and an unlabeled problem sample set;
carrying out labeling description on the untagged problem sample set to obtain a labeled problem sample set;
expanding the normal sample set and the labeled problem sample set to obtain an expanded sample set;
and placing the extended sample set into the original radiation source signal sample set to obtain a new radiation source signal sample set.
In a second aspect, the present application also provides a deep network-based radiation source individual analysis apparatus, the apparatus comprising:
an acquisition module for acquiring an original radiation source signal sample set comprising a plurality of radiation source signal samples;
the computing module is used for carrying out sample characteristic labeling on the original radiation source signal sample set after Grad-CAM computation to obtain a high-reaction labeling area of each radiation source signal sample;
the first processing module is used for analyzing the high-reaction labeling area of each radiation source signal sample to obtain a normal sample set and an unlabeled problem sample set;
the second processing module is used for carrying out labeling description on the untagged problem sample set to obtain a labeled problem sample set;
the third processing module is used for expanding the normal sample set and the labeling problem sample set to obtain an expanded sample set;
and a fourth processing module, configured to put the extended sample set into the original radiation source signal sample set, to obtain a new radiation source signal sample set.
The beneficial effects of the invention are as follows:
the invention carries out reverse reasoning according to the layering combination form of the neurons and understands the characteristic meaning of each stage, and provides a deep network visualization method (Grad-CAM) based on gradient positioning, which can carry out visualization presentation on a deep learning network under the condition of not modifying a model and retraining; in addition, the invention can analyze the behavior of the radiation source individual identification model, judge the uncertainty of the model and realize the interpretable expression of the radiation source individual. In the aspect of sample set processing, the invention can also expand and strengthen the sample set.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the embodiments of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims thereof as well as the appended drawings.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments will be briefly described below, it being understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and other related drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of an individual analysis method of a radiation source based on a deep network according to an embodiment of the invention;
FIG. 2 is a schematic diagram of an individual analysis device of a radiation source based on a deep network according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a second processing module according to an embodiment of the present invention.
The marks in the figure:
901. an acquisition module; 902. a computing module; 903. a first processing module; 904. a second processing module; 905. a third processing module; 906. a fourth processing module; 9031. a first judgment unit; 9032. a second judgment unit; 9033. a first processing unit; 9034. a second processing unit; 9041. a first analysis unit; 9042. a second analysis unit; 9043. a first calculation unit; 9044. a second calculation unit; 9045. a third judgment unit; 9046. a third calculation unit; 9047. a fourth calculation unit; 9048. a third analysis unit; 9049. a confusion analysis unit; 90491. a fifth calculation unit; 90492. a sixth calculation unit; 90493. a fourth analysis unit; 9051. a seventh calculation unit; 9052. an eighth calculation unit; 9053. a ninth calculation unit; 9054. a fifth analysis unit; 9055. a sixth analysis unit; 9056. and a seventh analysis unit.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. The components of the 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 invention, as 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 made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures. Meanwhile, in the description of the present invention, the terms "first", "second", and the like are used only to distinguish the description, and are not to be construed as indicating or implying relative importance.
Example 1:
the embodiment provides a radiation source individual analysis method based on a deep network.
Referring to fig. 1, the method is shown to include steps S1-S6, specifically:
s1, acquiring an original radiation source signal sample set, wherein the original radiation source signal sample set comprises a plurality of radiation source signal samples;
in step S1, for each radiation source signal sample, the collected radiation source signal sample is in the form of a sample point time domain amplitude since the signal parameters transmitted by the radiation source are the same each time. And marking the time domain signal samples with the same sampling point length according to labels such as the individual number of the radiation source, a training set, a test set, a signal to noise ratio and the like, wherein each radiation source signal sample comprises a signal section and a noise section.
S2, carrying out sample feature labeling on the original radiation source signal sample set through Grad-CAM calculation to obtain a high-reaction labeling area of each radiation source signal sample;
in step S2, the high-reaction labeling area refers to an area of higher importance or significance in the signal. These areas may contain critical information related to the task and can attract attention or attention of the model. The invention uses Grad-CAM algorithm to calculate the thermal value corresponding to the whole pulse signal at each importance point, and the larger the thermal value is, the more important the classification algorithm is, namely the high-response labeling area is corresponding.
In the present invention, the highly reactive labeling zone functions as: by determining the high-reaction labeling area, the method can find the area of the signal fingerprint characteristics, thereby improving the interpretation of the model. In addition, the analysis of the high-response labeling area can help to determine the extraction method of the signal fingerprint characteristics, optimize the signal processing algorithm and improve the identification accuracy and the robustness of the radiation source and the modulation signal.
In step S2, grad-CAM is calculated as follows:
in the above-mentioned method, the step of,representing the attention matrix of the model to the current input sample with respect to category c by the ReLU function; />Representing a certain characteristic layer, which corresponds to the characteristic layer output by the last convolution layer; />Representing feature layer->Middle (f)' ShutongA lane; />Representing the weight; />Indicating a total number of channels of 1 +.>Indicating a total number of channels of m.
in the above-mentioned method, the step of,representing the weight; />Representing the width +.>Height of the steel plate; />Representing the number of upper pixels of the picture;representing the number of pixels on the wide side, representing 1 pixel on the high side of the picture, and +.>The number of upper pixels representing the picture is H; />The number of pixels on the width of the picture is 1, < >>Representing the number of pixels over the width of a pictureW; />Representing the network +.>Predictive score,/->Representing feature layer->In the channel->,/>And coordinates are data at locations.
Firstly, establishing a network model to obtain a characteristic layer and a network predicted value, then carrying out directional propagation on the predicted value to obtain gradient information of an anti-return characteristic layer, calculating to obtain importance degree of each channel of the characteristic layer, then carrying out weighted summation, activating through a ReLU function, and finally obtaining a network attention area to realize interpretation of a classification basis and signal characteristic presentation of a deep neural network model in a thermodynamic diagram mode.
S3, analyzing the high-reaction labeling area of each radiation source signal sample to obtain a normal sample set and an unlabeled problem sample set;
in step S3, the normal sample set includes a first data set, a second data set, a third data set, and a fourth data set, and the unlabeled problem sample set includes a first abnormal sample set, a second abnormal sample set, a misidentification sample set, and a pollution sample set, including steps S31-S34, specifically including:
s31, judging the high-reaction labeling areas of the signal samples of each radiation source when the signal sample sets of the original radiation source are different samples of the same radiation source, and obtaining a first data set and a first abnormal sample set, wherein the first data set is a sample set with the same high-reaction labeling area of different samples of the same radiation source, and the first abnormal sample set is a sample set with different high-reaction labeling areas of different samples of the same radiation source;
in step S31, further, the capability features of different deep web learning models may be tested, specifically: aiming at the same sample set, testing whether the characteristic high-reflection areas identified by different deep learning models are the same, and if so, indicating that the radiation source has a recognized characteristic area; if it is different, it is stated that different deep learning models have feature selectivity.
S32, judging the high-reaction labeling area of each radiation source signal sample when different samples of different radiation sources are in the original radiation source signal sample set, and obtaining a second data set and a second abnormal sample set, wherein the second data set is a sample set with different high-reaction labeling areas of different samples of different radiation sources, and the second abnormal sample set is a sample set with the same high-reaction labeling areas of different samples of different radiation sources;
in step S32, further, the same deep learning model can be tested for its ability to identify different radiation sources. The method specifically comprises the following steps: aiming at samples of different radiation sources, analysis and identification accuracy and high-reflection area display results show that the lower the accuracy is, the less the high-reflection area is fixed, the greater the difficulty of feature extraction is.
S33, carrying out covering or noise adding treatment on the high-reaction marked area of each radiation source signal sample to obtain a third data set and a false identification sample set, wherein the third data set is a non-sensitive sample for covering or noise adding treatment, and the false identification sample set is a sensitive sample for covering or noise adding treatment;
the masking treatment: in step S33, after testing the masked high-response labeling area (i.e., zero-filling the data), re-identifying, and if the accuracy rate is obviously reduced, determining that the high-response labeling area is found; on the other hand, if the accuracy is not obviously reduced, whether the secondary high reaction labeling area is not covered or not can be judged, so that the characteristics are still obvious;
the noise adding process comprises the following steps: in step S33, the high-response labeling area may be subjected to noise adding processing, and then the recognition may be performed again; the influence of the noise size on the high-response labeling area can be observed, and the response of the model to the environmental noise or interference can be judged.
In the method, step S33 is set, the key feature position of the tested object is rapidly positioned through time domain shielding and identification test of the original sample, and the interpretability of the individual features of the radiation source is improved based on the result difference; in addition, through the comparison test of the generated sample and the original sample, the correlation influence of the sample on the generated model and the identification model is verified, and the influence of the characteristic shielding and the channel variation on the identification result is revealed.
And S34, carrying out noise section analysis on the high-reaction labeling area of each radiation source signal sample to obtain a fourth data set and a pollution sample set, wherein the fourth data set is a sample set of which the high-reaction labeling area of each radiation source signal sample does not contain a noise section, and the pollution sample set is a sample set of which the high-reaction labeling area of each radiation source signal sample contains a noise section.
In step S34, it is analyzed whether the high reaction labeling area is a noise segment. If so, the model learns the noise characteristics and does not learn the real radiation source characteristics. The model is greatly influenced by environment, has difficult mobility and needs to be corrected and perfected.
S4, carrying out labeling description on the untagged problem sample set to obtain a labeled problem sample set;
in step S4, the labeled problem sample set includes a time series label, where the time series label includes a first time label, a second time label, and a third time label, and step S4 includes step S41-step S45, specifically includes:
s41, carrying out acquisition time sequence analysis on the first abnormal sample set to obtain a first time tag;
s42, carrying out acquisition time sequence analysis on the second abnormal sample set to obtain a second time tag;
in step S41 and step S42, since the collected samples have the collection time labels or the serial numbers based on the collection time, the corresponding time labels can be obtained based on the time series analysis.
S43, carrying out probability calculation on the false recognition sample set through a preset formula to obtain a false recognition sample probability set;
in step S43, it is assumed that the existing data is M pulses arranged in time sequence in arrays belonging to different categoriesEvery pulse +.>Is a signal to be identified. Specifically, assume that the sample set is a time series of 10 sets of M pulses arranged in time order +.>Wherein->And the first five groups are from category 1 and the last five groups are from category 2.
Probability calculation: each pulse is provided withInputting into a trained neural network model, and finally outputting a probability output of the model through a softmax function>Wherein N represents the N-th class, < >>Representative pulseProbabilities belonging to N categories.
When probability calculation is carried out, the preset formula of the method is as follows:
in the above-mentioned method, the step of,representing a trained neural network model, +.>Representing a time sequence,/->M pulses representing an array of different categories arranged in time sequence, N representing the N-th category,/->Representing pulse->Probabilities belonging to N categories.
In the method, for each time seriesThe probability of belonging to each category can be calculated by a preset formula.
In step S43, by calculating the probability of the misrecognized sample, it is possible to determine the influence of the environment on the misrecognized sample, that is, if the misrecognized sample continuously appears, it is possible that the samples are all contaminated or are affected by the environment in the period of time; if the misrecognized samples occur randomly, the effects of possible environmental disturbances, etc., are more random.
S44, performing Euclidean distance calculation on every two false identification samples in the false identification sample probability set to obtain a similarity value of each false identification sample;
in step S44, euclidean distance calculation is performed on every two misidentified samples in the misidentified sample probability set, so as to further analyze whether there is a significant difference between each category and each category. The euclidean distance is prior art and will not be explained in detail herein.
And S45, judging a high-reaction labeling area according to the similarity value correspondence of each misidentification sample, and obtaining a third time tag.
In step S45, a similarity threshold may be preset, which is used to analyze whether the high reaction regions of the respective misrecognized samples are similar by performing a judgment of the high reaction labeling region corresponding to the similarity value of each misrecognized sample, and obtain the cause of misrecognized from the time dimension.
In step S4, the labeled problem sample set further includes a spatial distribution label, where the spatial distribution label includes a first spatial label, and step S4 includes steps S46-S48, specifically includes:
s46, inputting each false identification sample in the false identification sample set into a preset neural network model to obtain a prediction category of each false identification sample;
s47, carrying out thermal value sequence calculation through a Grad-CAM algorithm according to the prediction category of each misidentification sample to obtain a correctly classified sample thermal value sequence set and a misidentification sample thermal value sequence set;
in step S47, the prediction category of each misidentified sample is compared with the real label to obtain a correctly classified sample set and a misclassified sample set;
and respectively carrying out thermal value sequence calculation on the correctly classified sample set and the misclassified sample set through a Grad-CAM algorithm, and correspondingly obtaining the correctly classified sample thermal value sequence set and the misclassified sample thermal value sequence set.
S48, carrying out space analysis on the high-reaction labeling area according to the correctly classified sample thermal value sequence set and the misidentified sample thermal value sequence set to obtain a first space label.
In step S48, a similarity calculation is performed according to the correctly classified sample thermal value sequence set and the thermal value sequence set corresponding to the real tag, so as to obtain a first similarity mean value, where the similarity calculation method may use a euclidean distance.
And carrying out similarity calculation according to the misidentification sample thermal value sequence set and the thermal value sequence set corresponding to the misjudged label to obtain a second similarity mean value, wherein the similarity calculation method can adopt Euclidean distance.
In the method, when the first similarity mean value is smaller than the second similarity mean value, the misclassified sample is more similar to the high-reaction labeling area of the misclassified class set, so that the first space label is obtained.
In this method, further, spatial analysis of the high-reaction labeling area may be performed on the first abnormal sample set and the second abnormal sample set, that is, the spatial distribution label further includes a second spatial label and a third spatial label, which specifically includes:
performing spatial analysis on the high-reaction labeling area on the first abnormal sample set to obtain a second spatial label;
performing spatial analysis on the high-reaction labeling area on the second abnormal sample set to obtain a third spatial label;
the principle of the analysis method of the second space tag and the third space tag is the same as that of the analysis method of the first space tag.
In step S4, the labeled problem sample set further includes confusion labels, where step S4 includes step S49, and step S49 includes steps S491-S493, specifically including:
s491, inputting each problem sample in the unlabeled problem sample set into a preset neural network model to obtain a prediction category of each problem sample;
s492, constructing and obtaining a confusion matrix according to the prediction category of each problem sample;
in step S492, an n×n 0 matrix may be constructed, each sample is traversed, the prediction result and the true label are used as indexes, and then the element of the corresponding confusion matrix is incremented by one, i.e., for the sample with the prediction class i and the true class j, the element of the ith row and the jth column of the confusion matrix is incremented by one.
S493, performing confusion analysis on each problem sample according to the confusion matrix to obtain confusion labels.
In step S493, a confusing analysis is performed on each problem sample according to the confusion matrix, so as to determine which class is easier to misjudge with each class, that is, find the class corresponding to each row of the confusion matrix except the sub-maximum value of the diagonal element, thereby obtaining the confusion label.
In the method, aiming at the problem of insufficient radiation source signal samples, in order to expand the samples, improve the expansion capacity and enhance the individual identification capacity of the samples, step S4 is followed by step S5, which specifically comprises the following steps:
s5, expanding the normal sample set and the labeled problem sample set to obtain an expanded sample set;
in step S5, the set of extended samples includes a first set of extended samples and a second set of extended samples, where S5 includes steps S51-S53, specifically:
s51, calculating the normal sample set and the labeled problem sample set through a generated countermeasure network to obtain a model generated signal sample;
s52, judging a model generated signal sample to obtain a first expansion set;
in step S52, the model generated signal sample is determined, if the determination result is accurate and there is a high-response labeling area similar to the original signal, it is indicated that the intelligent generated model learns the characteristics of the high-response labeling area, and the sample generating capability of the individual characteristics of the radiation source is provided, and the higher-quality signal sample generated by expansion can be used for supplementing the radiation source signal sample set, so as to obtain the first expansion set.
And S53, updating the modulation mode of the normal sample set and the labeling problem sample set to obtain a second expansion set.
In step S53, the analysis radiation source expands the characteristic high-reaction labeling area for transmitting the single-frequency signal and the linear frequency signal when adopting the new modulation mode under the same parameter conditions such as frequency, signal length and the like, so as to obtain a second expansion set.
In step S5, the extended sample set further includes a third extended set, a fourth extended set, and a fifth extended set, where S5 includes steps S54 to S56, specifically includes:
s54, analyzing the normal sample set and the labeled problem sample set according to a time sequence to obtain a third expansion set;
in step S54, the normal sample set and the labeled problem sample set are analyzed according to a time sequence, on one hand, whether the characteristic high-reaction labeling area changes with time is analyzed, and on the other hand, whether the characteristic high-reaction labeling area changes after a time span, whether the change is generated by a radiation source or is influenced by environment is analyzed, so as to obtain a third expansion set.
S55, analyzing the normal sample set and the labeled problem sample set according to the space distance to obtain a fourth expansion set;
in step S55, the normal sample set and the labeled problem sample set are analyzed according to the spatial distance, so as to analyze whether the characteristic high-reaction labeling area changes along with the spatial distance; on the other hand analysis of how large a spatial span will lead to a change in the high reflection area and if a change is made, analysis of whether the change is due to the generation of radiation sources or due to environmental influences is made to obtain a fourth set of extensions.
S56, analyzing the normal sample set and the labeling problem sample set according to the multi-frequency signal to obtain a fifth expansion set.
In step S56, the normal sample set and the labeled problem sample set are analyzed according to the multi-frequency signal, and the influence of the radiation source emission frequency expansion variation is analyzed. The radiation source emits multi-frequency signals (such as linear frequency modulation signals, spread spectrum signals based on sequences and the like) and the radiation source emits single-frequency signals are compared, whether the multi-frequency signals are added with characteristics or not is analyzed, and characteristic position differences of all frequency points are analyzed to obtain a fifth spread set.
S6, placing the expanded sample set into the original radiation source signal sample set to obtain a new radiation source signal sample set.
In step S6, the new radiation source signal sample set is obtained by radiation source individual analysis based on a deep network, sample feature labeling is carried out after Grad-CAM calculation, and a high-reaction labeling area of each radiation source signal sample is obtained, so that the visual presentation of the feature positions of individual samples is realized; in addition, the high-response labeling area is analyzed to obtain the behavior of the radiation source individual identification model, and the uncertainty of the model is judged, so that the interpretable expression of the radiation source individual is realized. Finally, the individual recognition capability of the sample is enhanced through expansion.
Example 2:
as shown in fig. 2, the present embodiment provides a deep network-based radiation source individual analysis device, which includes:
an acquisition module 901 for acquiring an original radiation source signal sample set, the original radiation source signal sample set comprising a plurality of radiation source signal samples;
the calculating module 902 is configured to perform sample feature labeling on the original radiation source signal sample set after Grad-CAM calculation, so as to obtain a high-reaction labeling area of each radiation source signal sample;
the first processing module 903 is configured to analyze the high-reaction labeling area of each radiation source signal sample to obtain a normal sample set and an unlabeled problem sample set;
a second processing module 904, configured to perform a labeling description on the untagged problem sample set to obtain a labeled problem sample set;
a third processing module 905, configured to expand the normal sample set and the labeled problem sample set to obtain an expanded sample set;
a fourth processing module 906 is configured to put the extended sample set into the original radiation source signal sample set to obtain a new radiation source signal sample set.
In the method disclosed in the present invention, when the normal sample set includes a first data set, a second data set, a third data set, and a fourth data set, the unlabeled problem sample set includes a first abnormal sample set, a second abnormal sample set, a false identification sample set, and a contaminated sample set in the first processing module 903, including:
the first judging unit 9031 is configured to judge, when the original radiation source signal sample set is a different sample of the same radiation source, a high-reaction labeling area of each radiation source signal sample to obtain a first data set and a first abnormal sample set, where the first data set is a sample set of different samples of the same radiation source in the high-reaction labeling area, and the first abnormal sample set is a sample set of different samples of the same radiation source in the high-reaction labeling area;
the second judging unit 9032 is configured to judge, when the original radiation source signal sample set is different samples of different radiation sources, a high-reaction labeling area of each radiation source signal sample to obtain a second data set and a second abnormal sample set, where the second data set is a sample set in which different samples of different radiation sources are the same in the high-reaction labeling area, and the second abnormal sample set is a sample set in which different samples of different radiation sources are different in the high-reaction labeling area;
a first processing unit 9033, configured to mask or denoise a high-response labeling area of each radiation source signal sample, to obtain a third data set and a false identification sample set, where the third data set is a sensitive sample for masking or denoise, and the false identification sample set is a non-sensitive sample for masking or denoise;
and the second processing unit 9034 is configured to perform noise segment analysis on the high-response labeling area of each radiation source signal sample to obtain a fourth data set and a pollution sample set, where the fourth data set is a sample set of each high-response labeling area of the radiation source signal sample including a noise segment, and the pollution sample set is a sample set of each high-response labeling area of the radiation source signal sample including no noise segment.
As shown in fig. 3, in one method disclosed in the present invention, in the second processing module 904, the labeled problem sample set includes a time series label, where the time series label includes a first time label, a second time label, and a third time label, and includes:
a first analysis unit 9041, configured to perform acquisition timing analysis on the first abnormal sample set, to obtain a first time tag;
a second analysis unit 9042, configured to perform acquisition timing analysis on the second abnormal sample set, to obtain a second time tag;
the first calculating unit 9043 is configured to perform probability calculation on the misrecognition sample set through a preset formula to obtain a misrecognition sample probability set;
the second calculating unit 9044 is configured to perform euclidean distance calculation on every two misidentified samples in the misidentified sample probability set, so as to obtain a similarity value of each misidentified sample;
and a third judging unit 9045, configured to correspondingly judge the high-reaction labeling area according to the similarity value of each misidentification sample, so as to obtain a third time tag.
In one method disclosed in the present invention, in the second processing module 904, the labeled problem sample set further includes a spatial distribution label, where the spatial distribution label includes a first spatial label, and includes:
a third computing unit 9046, configured to input each misidentification sample in the misidentification sample set into a preset neural network model, to obtain a prediction category of each misidentification sample;
a fourth calculation unit 9047, configured to calculate a thermal value sequence according to the prediction category of each misidentified sample through a Grad-CAM algorithm, so as to obtain a correctly classified sample thermal value sequence set and a misidentified sample thermal value sequence set;
and the third analysis unit 9048 is configured to perform spatial analysis of the high-reaction labeling area according to the correctly classified sample thermal value sequence set and the misidentified sample thermal value sequence set, so as to obtain a first spatial label.
In one method disclosed in the present invention, in the second processing module 904, a confusion analysis unit 9049 is further included, where the confusion analysis unit 9049 includes:
a fifth computing unit 90491, configured to input each problem sample in the unlabeled problem sample set into a preset neural network model, to obtain a prediction category of each problem sample;
a sixth calculation unit 90492, configured to construct an confusion matrix according to the prediction type of each problem sample;
and a fourth analysis unit 90493, configured to perform a confusion analysis on each problem sample according to the confusion matrix, to obtain a confusion label.
In one disclosed method, in the third processing module 905, the set of expanded samples includes a first set of expanded samples and a second set of expanded samples, including:
a seventh calculating unit 9051, configured to calculate the normal sample set and the labeled problem sample set by generating an countermeasure network, to obtain a model generated signal sample;
an eighth calculating unit 9052, configured to determine the model generated signal sample, to obtain a first extension set;
and a ninth calculating unit 9053, configured to update the modulation modes of the normal sample set and the labeled problem sample set, to obtain a second extension set.
In one disclosed method, in the third processing module 905, the extended sample set further includes a third extended set, a fourth extended set, and a fifth extended set, including:
a fifth analysis unit 9054, configured to analyze the normal sample set and the labeled problem sample set according to a time sequence, to obtain a third extension set;
a sixth analysis unit 9055, configured to analyze the normal sample set and the labeled problem sample set according to a spatial distance, to obtain a fourth extension set;
and a seventh analysis unit 9056, configured to analyze the normal sample set and the labeled problem sample set according to the multi-frequency signal, to obtain a fifth extension set.
It should be noted that, regarding the apparatus in the above embodiments, the specific manner in which the respective modules perform the operations has been described in detail in the embodiments regarding the method, and will not be described in detail herein.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.
Claims (10)
1. A deep network-based radiation source individual analysis method, comprising:
acquiring an original radiation source signal sample set, the original radiation source signal sample set comprising a plurality of radiation source signal samples;
carrying out sample characteristic labeling on the original radiation source signal sample set through Grad-CAM calculation to obtain a high-reaction labeling area of each radiation source signal sample;
analyzing the high-reaction labeling area of each radiation source signal sample to obtain a normal sample set and an unlabeled problem sample set;
carrying out labeling description on the untagged problem sample set to obtain a labeled problem sample set;
expanding the normal sample set and the labeled problem sample set to obtain an expanded sample set;
and placing the extended sample set into the original radiation source signal sample set to obtain a new radiation source signal sample set.
2. The deep network-based radiation source individual analysis method of claim 1, wherein analyzing the highly reactive labeling area of each of the radiation source signal samples results in a normal sample set and an unlabeled problem sample set, the normal sample set including a first data set, a second data set, a third data set, and a fourth data set, the unlabeled problem sample set including a first abnormal sample set, a second abnormal sample set, a misrecognized sample set, and a contaminated sample set, comprising:
when the original radiation source signal sample sets are different samples of the same radiation source, judging a high-reaction labeling area of each radiation source signal sample to obtain a first data set and a first abnormal sample set, wherein the first data set is a sample set, in which different samples of the same radiation source are in the same high-reaction labeling area, and the first abnormal sample set is a sample set, in which different samples of the same radiation source are in different high-reaction labeling areas;
when the original radiation source signal sample set is different samples of different radiation sources, judging a high-reaction labeling area of each radiation source signal sample to obtain a second data set and a second abnormal sample set, wherein the second data set is a sample set of different samples of different radiation sources in the high-reaction labeling area, and the second abnormal sample set is a sample set of different samples of different radiation sources in the high-reaction labeling area;
masking or denoising the high-reaction labeling area of each radiation source signal sample to obtain a third data set and a false identification sample set, wherein the third data set is a non-sensitive sample for masking or denoising, and the false identification sample set is a sensitive sample for masking or denoising;
and carrying out noise section analysis on the high-reaction labeling area of each radiation source signal sample to obtain a fourth data set and a pollution sample set, wherein the fourth data set is a sample set of which the high-reaction labeling area of each radiation source signal sample does not contain a noise section, and the pollution sample set is a sample set of which the high-reaction labeling area of each radiation source signal sample contains a noise section.
3. The deep network-based radiation source individual analysis method of claim 2, wherein the labeling of the unlabeled question sample set to obtain a labeled question sample set, the labeled question sample set including a time series label, the time series label including a first time label, a second time label, and a third time label, comprises:
carrying out acquisition time sequence analysis on the first abnormal sample set to obtain a first time tag;
performing acquisition time sequence analysis on the second abnormal sample set to obtain a second time tag;
probability calculation is carried out on the false recognition sample set through a preset formula, and a false recognition sample probability set is obtained;
performing Euclidean distance calculation on every two misidentification samples in the misidentification sample probability set to obtain a similarity value of each misidentification sample;
and judging the high-reaction labeling area according to the similarity value correspondence of each misidentification sample, and obtaining a third time label.
4. The deep network-based radiation source individual analysis method of claim 2, wherein the unlabeled question sample set is subjected to labeling description to obtain a labeled question sample set, the labeled question sample set further comprises a spatial distribution label, the spatial distribution label comprises a first spatial label, and the method comprises the steps of:
inputting each false identification sample in the false identification sample set into a preset neural network model to obtain a prediction category of each false identification sample;
carrying out thermal value sequence calculation through Grad-CAM algorithm according to the prediction category of each misidentification sample to obtain a correctly classified sample thermal value sequence set and a misidentification sample thermal value sequence set;
and carrying out space analysis on the high-reaction labeling area according to the correctly classified sample thermal value sequence set and the misidentified sample thermal value sequence set to obtain a first space label.
5. The deep network-based radiation source individual analysis method of claim 1, wherein expanding the normal sample set and the tagged problem sample set to obtain an expanded sample set, the expanded sample set comprising a first expanded set and a second expanded set, comprises:
calculating the normal sample set and the labeled problem sample set through a generated countermeasure network to obtain a model generated signal sample;
judging the model generated signal sample to obtain a first expansion set;
and carrying out modulation mode updating on the normal sample set and the labeling problem sample set to obtain a second expansion set.
6. A deep network-based radiation source individual analysis device, comprising:
an acquisition module for acquiring an original radiation source signal sample set comprising a plurality of radiation source signal samples;
the computing module is used for carrying out sample characteristic labeling on the original radiation source signal sample set after Grad-CAM computation to obtain a high-reaction labeling area of each radiation source signal sample;
the first processing module is used for analyzing the high-reaction labeling area of each radiation source signal sample to obtain a normal sample set and an unlabeled problem sample set;
the second processing module is used for carrying out labeling description on the untagged problem sample set to obtain a labeled problem sample set;
the third processing module is used for expanding the normal sample set and the labeling problem sample set to obtain an expanded sample set;
and a fourth processing module, configured to put the extended sample set into the original radiation source signal sample set, to obtain a new radiation source signal sample set.
7. The deep network-based radiation source individual analysis device defined in claim 6, wherein, in the first processing module, when the normal sample set comprises a first data set, a second data set, a third data set, and a fourth data set, the unlabeled problem sample set comprises a first abnormal sample set, a second abnormal sample set, a misrecognized sample set, and a contaminated sample set, comprising:
the first judging unit is used for judging the high-reaction labeling area of each radiation source signal sample when the original radiation source signal sample set is different samples of the same radiation source, so as to obtain a first data set and a first abnormal sample set, wherein the first data set is a sample set, in which different samples of the same radiation source are in the high-reaction labeling area, and the first abnormal sample set is a sample set, in which different samples of the same radiation source are in the high-reaction labeling area;
the second judging unit is used for judging the high-reaction labeling area of each radiation source signal sample when the original radiation source signal sample set is different samples of different radiation sources, so as to obtain a second data set and a second abnormal sample set, wherein the second data set is a sample set of different samples of different radiation sources in the high-reaction labeling area, and the second abnormal sample set is a sample set of different samples of different radiation sources in the high-reaction labeling area;
the first processing unit is used for carrying out covering or noise adding processing on the high-reaction marked area of each radiation source signal sample to obtain a third data set and a false identification sample set, wherein the third data set is a non-sensitive sample for covering or noise adding processing, and the false identification sample set is a sensitive sample for covering or noise adding processing;
and the second processing unit is used for carrying out noise section analysis on the high-reaction labeling area of each radiation source signal sample to obtain a fourth data set and a pollution sample set, wherein the fourth data set is a sample set of which the high-reaction labeling area of each radiation source signal sample does not contain a noise section, and the pollution sample set is a sample set of which the high-reaction labeling area of each radiation source signal sample contains a noise section.
8. The deep network-based radiation source individual analysis device defined in claim 7, wherein in the second processing module, the labeled problem sample set comprises a time series label comprising a first time label, a second time label, and a third time label, comprising:
the first analysis unit is used for carrying out acquisition time sequence analysis on the first abnormal sample set to obtain a first time tag;
the second analysis unit is used for carrying out acquisition time sequence analysis on the second abnormal sample set to obtain a second time tag;
the first calculation unit is used for carrying out probability calculation on the false identification sample set through a preset formula to obtain a false identification sample probability set;
the second calculation unit is used for carrying out Euclidean distance calculation on every two false identification samples in the false identification sample probability set to obtain a similarity value of each false identification sample;
and the third judging unit is used for correspondingly judging the high-reaction labeling area according to the similarity value of each false identification sample to obtain a third time tag.
9. The deep network-based radiation source individual analysis device defined in claim 7, wherein in the second processing module, the labeled problem sample set further comprises a spatial distribution label comprising a first spatial label comprising:
the third calculation unit is used for inputting each false identification sample in the false identification sample set into a preset neural network model to obtain a prediction category of each false identification sample;
the fourth calculation unit is used for calculating a thermal value sequence through a Grad-CAM algorithm according to the prediction category of each misidentification sample to obtain a correctly classified sample thermal value sequence set and a misidentification sample thermal value sequence set;
and the third analysis unit is used for carrying out space analysis on the high-reaction labeling area according to the correct classification sample thermal value sequence set and the misidentification sample thermal value sequence set to obtain a first space label.
10. The deep network-based radiation source individual analysis device defined in claim 6, wherein in the third processing module, the extended sample set comprises a first extended set and a second extended set comprising:
a seventh calculation unit, configured to calculate the normal sample set and the labeled problem sample set by generating an countermeasure network, to obtain a model generated signal sample;
an eighth calculation unit, configured to determine a signal sample generated by the model, to obtain a first extension set;
and a ninth calculation unit, configured to update the modulation mode of the normal sample set and the labeled problem sample set, so as to obtain a second extension set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310675856.8A CN116401588B (en) | 2023-06-08 | 2023-06-08 | Radiation source individual analysis method and device based on deep network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310675856.8A CN116401588B (en) | 2023-06-08 | 2023-06-08 | Radiation source individual analysis method and device based on deep network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116401588A true CN116401588A (en) | 2023-07-07 |
CN116401588B CN116401588B (en) | 2023-08-15 |
Family
ID=87014664
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310675856.8A Active CN116401588B (en) | 2023-06-08 | 2023-06-08 | Radiation source individual analysis method and device based on deep network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116401588B (en) |
Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111897985A (en) * | 2020-06-23 | 2020-11-06 | 西安交通大学 | Image multi-label classification method, system, equipment and readable storage medium |
CN112308008A (en) * | 2020-11-12 | 2021-02-02 | 电子科技大学 | Radar radiation source individual identification method based on working mode open set of transfer learning |
CN112613575A (en) * | 2020-12-30 | 2021-04-06 | 清华大学 | Data set expansion method, training method and device of image classification model |
CN114019467A (en) * | 2021-10-25 | 2022-02-08 | 哈尔滨工程大学 | Radar signal identification and positioning method based on MobileNet model transfer learning |
US20220217619A1 (en) * | 2021-01-06 | 2022-07-07 | Bae Systems Information And Electronic Systems Integration Inc. | Artificial intelligence radio classifier and identifier |
CN114912482A (en) * | 2022-04-30 | 2022-08-16 | 中国人民解放军海军航空大学 | Method and device for identifying radiation source |
CN115081487A (en) * | 2022-07-08 | 2022-09-20 | 电子科技大学 | Cross-time radiation source individual identification method based on multi-domain transfer learning |
CN115270878A (en) * | 2022-07-28 | 2022-11-01 | 哈尔滨工业大学 | Radiation source individual identification method based on class activation diagram and SincNet network |
CN115270872A (en) * | 2022-07-26 | 2022-11-01 | 中山大学 | Radar radiation source individual small sample learning and identifying method, system, device and medium |
WO2022268618A1 (en) * | 2021-06-24 | 2022-12-29 | Koninklijke Philips N.V. | Multi-energy x-ray imaging with anatomical intelligence |
CN115712856A (en) * | 2022-11-30 | 2023-02-24 | 厦门大学 | CNN-CBAM-BDA-based cross-over receiver radiation source identification method and system |
CN115761225A (en) * | 2022-11-03 | 2023-03-07 | 北京科技大学 | Image annotation method based on neural network interpretability |
CN115829909A (en) * | 2022-05-27 | 2023-03-21 | 华侨大学 | Forgery detection method based on feature enhancement and spectrum analysis |
-
2023
- 2023-06-08 CN CN202310675856.8A patent/CN116401588B/en active Active
Patent Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111897985A (en) * | 2020-06-23 | 2020-11-06 | 西安交通大学 | Image multi-label classification method, system, equipment and readable storage medium |
CN112308008A (en) * | 2020-11-12 | 2021-02-02 | 电子科技大学 | Radar radiation source individual identification method based on working mode open set of transfer learning |
CN112613575A (en) * | 2020-12-30 | 2021-04-06 | 清华大学 | Data set expansion method, training method and device of image classification model |
US20220217619A1 (en) * | 2021-01-06 | 2022-07-07 | Bae Systems Information And Electronic Systems Integration Inc. | Artificial intelligence radio classifier and identifier |
WO2022268618A1 (en) * | 2021-06-24 | 2022-12-29 | Koninklijke Philips N.V. | Multi-energy x-ray imaging with anatomical intelligence |
CN114019467A (en) * | 2021-10-25 | 2022-02-08 | 哈尔滨工程大学 | Radar signal identification and positioning method based on MobileNet model transfer learning |
CN114912482A (en) * | 2022-04-30 | 2022-08-16 | 中国人民解放军海军航空大学 | Method and device for identifying radiation source |
CN115829909A (en) * | 2022-05-27 | 2023-03-21 | 华侨大学 | Forgery detection method based on feature enhancement and spectrum analysis |
CN115081487A (en) * | 2022-07-08 | 2022-09-20 | 电子科技大学 | Cross-time radiation source individual identification method based on multi-domain transfer learning |
CN115270872A (en) * | 2022-07-26 | 2022-11-01 | 中山大学 | Radar radiation source individual small sample learning and identifying method, system, device and medium |
CN115270878A (en) * | 2022-07-28 | 2022-11-01 | 哈尔滨工业大学 | Radiation source individual identification method based on class activation diagram and SincNet network |
CN115761225A (en) * | 2022-11-03 | 2023-03-07 | 北京科技大学 | Image annotation method based on neural network interpretability |
CN115712856A (en) * | 2022-11-30 | 2023-02-24 | 厦门大学 | CNN-CBAM-BDA-based cross-over receiver radiation source identification method and system |
Non-Patent Citations (4)
Title |
---|
MINGZHE ZHU 等: "A probe-feature for specific emitter identification using axiom-based grad-CAM", 《SIGNAL PROCESSING》, vol. 201, pages 1 - 8 * |
刘文斌 等: "辐射源个体识别的一种可解释性测试架构", 《太赫兹科学与电子信息学报》, vol. 21, no. 06, pages 734 - 744 * |
古有志: "基于深度学习的雷达辐射源个体识别技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》, no. 2021, pages 136 - 1435 * |
林子榆 等: "基于深度自编码器的辐射源个体开集识别", 《太赫兹科学与电子信息学报》, vol. 20, no. 12, pages 1285 - 1291 * |
Also Published As
Publication number | Publication date |
---|---|
CN116401588B (en) | 2023-08-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111444790B (en) | Pulse-level intelligent identification method for multifunctional radar working mode sequence | |
CN115618296B (en) | Dam monitoring time sequence data anomaly detection method based on graph attention network | |
CN114509266B (en) | Bearing health monitoring method based on fault feature fusion | |
CN104766099A (en) | Distributed optical fiber circumference vibration signal processing and recognizing method based on image | |
CN111983569B (en) | Radar interference suppression method based on neural network | |
CN116051479A (en) | Textile defect identification method integrating cross-domain migration and anomaly detection | |
Ye et al. | A deep learning-based method for automatic abnormal data detection: Case study for bridge structural health monitoring | |
CN112084860A (en) | Target object detection method and device and thermal power plant detection method and device | |
Pucci et al. | Human activities classification using biaxial seismic sensors | |
CN114510960A (en) | Method for recognizing distributed optical fiber sensor system mode | |
CN111753877A (en) | Product quality detection method based on deep neural network transfer learning | |
CN116401588B (en) | Radiation source individual analysis method and device based on deep network | |
CN108960005B (en) | Method and system for establishing and displaying object visual label in intelligent visual Internet of things | |
Sony | Towards multiclass damage detection and localization using limited vibration measurements | |
Lee et al. | Neuralfp: out-of-distribution detection using fingerprints of neural networks | |
CN112966710A (en) | FY-3D infrared hyperspectral cloud detection method based on linear discriminant analysis | |
CN117041972A (en) | Channel-space-time attention self-coding based anomaly detection method for vehicle networking sensor | |
CN115452957B (en) | Small sample metal damage identification method based on attention prototype network | |
Yokkampon et al. | Autoencoder with Spiking in Frequency Domain for Anomaly Detection of Uncertainty Event. | |
Mansouri et al. | Enhanced monitoring of environmental processes | |
Yang et al. | Predicting the suitability for scene matching using SVM | |
Priyadharsini et al. | Performance Investigation of Handwritten Equation Solver using CNN for Betterment | |
Kühne et al. | Defending against adversarial attacks on time-series with selective classification | |
Choi et al. | Self-Supervised Time-Series Anomaly Detection Using Learnable Data Augmentation | |
Niemistö et al. | Correction of misclassifications using a proximity-based estimation method |
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 |