CN113762333B - Unsupervised anomaly detection method and system based on double-flow joint density estimation - Google Patents

Unsupervised anomaly detection method and system based on double-flow joint density estimation Download PDF

Info

Publication number
CN113762333B
CN113762333B CN202110817301.3A CN202110817301A CN113762333B CN 113762333 B CN113762333 B CN 113762333B CN 202110817301 A CN202110817301 A CN 202110817301A CN 113762333 B CN113762333 B CN 113762333B
Authority
CN
China
Prior art keywords
network
coding
density estimation
dimensional self
joint density
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.)
Active
Application number
CN202110817301.3A
Other languages
Chinese (zh)
Other versions
CN113762333A (en
Inventor
刘忆森
周松斌
邱泽帆
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.)
Institute of Intelligent Manufacturing of Guangdong Academy of Sciences
Original Assignee
Institute of Intelligent Manufacturing of Guangdong Academy of Sciences
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 Institute of Intelligent Manufacturing of Guangdong Academy of Sciences filed Critical Institute of Intelligent Manufacturing of Guangdong Academy of Sciences
Priority to CN202110817301.3A priority Critical patent/CN113762333B/en
Publication of CN113762333A publication Critical patent/CN113762333A/en
Application granted granted Critical
Publication of CN113762333B publication Critical patent/CN113762333B/en
Active 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
    • G06F18/21—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
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Image Analysis (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)

Abstract

The embodiment of the invention provides an unsupervised anomaly detection method and system based on double-current joint density estimation, which adopts a one-dimensional self-coding network to extract average spectral characteristics, adopts a two-dimensional self-coding network to extract principal component image spatial characteristics, and then performs fusion and density estimation on double-current characteristics of a normal sample through a joint density estimation network; and designing a balance penalty loss function to balance the contribution balance of the spectral feature and the spatial feature and prevent a certain branch from being over-fitted. The method can realize unsupervised spectrum imaging data anomaly detection based on 'space spectrum information' combination, and the anomaly identification effect is superior to that of the traditional self-coding reconstruction method.

Description

Unsupervised anomaly detection method and system based on double-flow joint density estimation
Technical Field
The embodiment of the invention relates to the technical field of spectral imaging data analysis, in particular to an unsupervised anomaly detection method and system based on double-flow joint density estimation.
Background
The spectrum imaging sensing technology such as near-infrared hyperspectrum, raman imaging, fluorescence spectrum imaging and terahertz imaging has wide application scenes, including food adulteration detection, agricultural product quality detection, drug component analysis, fake drug identification, microorganism content detection, oil gas component analysis and the like. In many quality control application scenes based on spectral imaging, the problems of various quality defect types, small quantity of quality abnormal samples, difficulty in collection and the like exist. Therefore, it is very difficult to actually control the quality by training a classification model by collecting a large number of each type of abnormal sample. For example, in the field of food and drug adulteration, the problems of rare adulteration samples, multiple adulterants and the possibility of multiple adulterants exist simultaneously exist. Therefore, only the normal/qualified samples which are conveniently obtained are adopted for carrying out unsupervised training, and various abnormal or defective samples are identified, so that effective quality control is realized, and the method has important practical application significance.
In addition, in the nondestructive testing based on spectral imaging, the spatial information of the spectral imaging data is not fully utilized, and most of the algorithms still adopt the average spectrum of the effective region for modeling at present.
Disclosure of Invention
The embodiment of the invention provides an unsupervised anomaly detection method and system based on double-current joint density estimation, which can realize unsupervised spectral imaging data anomaly detection based on 'space spectrum information' combination, and the anomaly identification effect is superior to that of the traditional self-encoding reconstruction method.
In a first aspect, an embodiment of the present invention provides an unsupervised anomaly detection method based on dual-stream joint density estimation, including:
s1, collecting spectral imaging data of a sample to be detected, and constructing a sample set based on the spectral imaging data;
s2, constructing a double-flow joint density estimation network, wherein the double-flow joint density estimation network comprises a one-dimensional self-coding network, a two-dimensional self-coding network and a joint density estimation network; the one-dimensional self-coding network is used for coding, decoding and reconstructing an average spectrum of spectral imaging data, the two-dimensional self-coding network is used for coding, decoding and reconstructing first n principal component PCA images of the spectral imaging data, and the joint density estimation network is used for carrying out double-flow feature fusion and density estimation on average spectral coding features obtained by the one-dimensional self-coding network and PCA space coding features obtained by the two-dimensional self-coding network;
s3, training the double-current joint density estimation network based on the sample set;
and S4, carrying out sample prediction based on the trained double-current joint density estimation network.
Preferably, the step S1 specifically includes:
acquiring first spectral imaging data of a normal sample and second spectral imaging data of an abnormal sample;
segmenting the first spectral imaging data based on a watershed algorithm to obtain effective pixels of each normal sample and each abnormal sample;
constructing a training sample set based on the first spectral imaging data, and constructing a test sample set based on the first spectral imaging data and the second spectral imaging data;
averaging the spectrums of all effective pixels in a training sample set and a testing sample set to obtain an average spectrum, and taking the average spectrum as the input of the one-dimensional self-coding network; and performing principal component analysis on all effective pixels in the training sample set, and performing data dimension reduction to obtain the first n principal component PCA images as the input of the two-dimensional self-coding network.
Preferably, after the dual-stream joint density estimation network is constructed in step S2, the method further includes:
constructing a loss function of the double-flow joint density estimation network:
L＝α 1 *L AE_1D +α 2 *L AE_2D +α 3 *L GMMnet +α 4 *L P
wherein L is AE_1D 、L AE_2D 、L GMMnet 、L P Respectively a one-dimensional self-coding network loss function, a two-dimensional self-coding network loss function, a joint density estimation network loss function, a balance penalty loss function, alpha 1 、α 2 、α 3 、α 4 To take on a value of [0,1]Constant coefficient of between.
Preferably, in step S2, the joint density estimation network includes m 1 +m 2 A fully connected layer, wherein the first m 1 The full connection layer is a feature fusion layer 2 Each full connection layer is a density estimation layer; the coding characteristics c obtained by the one-dimensional self-coding network 1D And coding characteristics c obtained from two-dimensional self-coding network 2D Cascading is carried out and used as the input of a characteristic fusion layer, and the obtained fusion characteristic c is used as the input of a density estimation layer; the output layer of the joint density estimation network comprises K nodes corresponding to K Gaussian partial models, and the output layer adopts a softmax function as nonlinear excitation.
Preferably, the one-dimensional self-coding network loss function L AE_1D Comprises the following steps:
Figure GDA0003991714870000031
wherein the content of the first and second substances,
Figure GDA0003991714870000032
the spectrum is averaged for the ith training sample,
Figure GDA0003991714870000033
obtaining a reconstructed spectrum for the ith training sample average spectrum after passing through a one-dimensional self-coding network;
the two-dimensional self-coding network loss function L AE_2D Comprises the following steps:
Figure GDA0003991714870000034
wherein, the first and the second end of the pipe are connected with each other,
Figure GDA0003991714870000035
for the PCA image of the ith training sample,
Figure GDA0003991714870000036
obtaining a reconstructed PCA image for the first n principal component PCA images of the ith training sample after passing through a two-dimensional self-coding network;
the joint density estimation network loss function L GMMnet Comprises the following steps:
Figure GDA0003991714870000037
wherein, c i The average spectral coding characteristics of the ith sample and the PCA spatial coding characteristics are fused to form a combined coding characteristic;
Figure GDA0003991714870000038
Σ k respectively predicting probability, mean value and covariance of the kth sub-model in the multivariate mixture Gaussian model;
the specific calculation method comprises the following steps:
Figure GDA0003991714870000039
Figure GDA00039917148700000310
Figure GDA00039917148700000311
the equilibrium penalty loss function L P Comprises the following steps:
Figure GDA00039917148700000312
wherein, w 1D,l Estimating c in network feature fusion layer for joint density 1D Corresponding network weight, w 2D,s Estimating c in network feature fusion layer for joint density 2D The corresponding network weight.
Preferably, the step S4 specifically includes:
inputting the prediction sample into a trained double-current joint density estimation network to obtain a likelihood function of the prediction sample;
and taking the likelihood function as an abnormal score, if the abnormal score is larger than a threshold th, judging that the prediction sample is abnormal, otherwise, judging that the prediction sample is normal.
Preferably, in step S4, a likelihood function E (c) of the sample is predicted t ) Comprises the following steps:
Figure GDA0003991714870000041
wherein, c t Obtaining joint coding characteristics for a prediction sample through a one-dimensional self-coding network and a two-dimensional self-coding network;
Figure GDA0003991714870000042
outputting a predicted value of a k Gaussian component model of a prediction sample, namely the k node output of the double-current joint density estimation network;
Figure GDA0003991714870000043
∑ k,prior respectively is the prior mean value and the prior covariance of the kth Gaussian mixture model in the multi-element mixed Gaussian model obtained by training.
In a second aspect, an embodiment of the present invention provides an unsupervised anomaly detection system based on dual-stream joint density estimation, including:
the sample acquisition module is used for acquiring spectral imaging data of a sample to be detected and constructing a sample set based on the spectral imaging data;
the network construction module is used for constructing a double-flow joint density estimation network, and the double-flow joint density estimation network comprises a one-dimensional self-coding network, a two-dimensional self-coding network and a joint density estimation network; the one-dimensional self-coding network is used for coding, decoding and reconstructing an average spectrum of spectral imaging data, the two-dimensional self-coding network is used for coding, decoding and reconstructing the first n principal component PCA images of the spectral imaging data, and the joint density estimation network is used for performing double-current feature fusion and density estimation on average spectral coding features obtained by the one-dimensional self-coding network and PCA space coding features obtained by the two-dimensional self-coding network;
the training module is used for training the double-current joint density estimation network based on the sample set;
and the prediction module is used for predicting samples based on the trained double-flow joint density estimation network.
In a third aspect, an embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor, when executing the program, implements the steps of the unsupervised anomaly detection method based on dual-stream joint density estimation according to the embodiment of the first aspect of the present invention.
In a fourth aspect, an embodiment of the present invention provides a non-transitory computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps of the unsupervised anomaly detection method based on dual-stream joint density estimation according to an embodiment of the first aspect of the present invention.
The embodiment of the invention provides an unsupervised anomaly detection method and system based on double-current joint density estimation, which adopts a one-dimensional self-coding network to extract average spectral characteristics, adopts a two-dimensional self-coding network to extract principal component image spatial characteristics, and then performs fusion and density estimation on double-current characteristics of a normal sample through a joint density estimation network; and designing a balance penalty loss function to balance the contribution balance of the spectral feature and the spatial feature and prevent a certain branch from being over-fitted. The method can realize unsupervised spectrum imaging data anomaly detection based on 'space spectrum information' combination, and the anomaly identification effect is superior to that of the traditional self-coding reconstruction method.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
FIG. 1 is a flow chart of an unsupervised anomaly detection method based on dual-stream joint density estimation according to an embodiment of the present invention;
FIG. 2 is a diagram of a dual-stream joint density estimation network architecture according to an embodiment of the present invention;
fig. 3 is a schematic physical structure diagram according to an embodiment of the present invention.
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. 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.
In the embodiment of the present application, the term "and/or" is only one kind of association relationship describing an associated object, and means that three relationships may exist, for example, a and/or B may mean: a exists alone, A and B exist simultaneously, and B exists alone.
The terms "first" and "second" in the embodiments of the present application are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present application, the terms "comprise" and "have", as well as any variations thereof, are intended to cover a non-exclusive inclusion. For example, a system, product or apparatus that comprises a list of elements or components is not limited to only those elements or components but may alternatively include other elements or components not expressly listed or inherent to such product or apparatus. In the description of the present application, "plurality" means at least two, e.g., two, three, etc., unless explicitly specifically limited otherwise.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.
In the nondestructive testing based on spectral imaging, the spatial information of spectral imaging data is not fully utilized, and most of the current algorithms still adopt the average spectrum of an effective area for modeling. In recent years, some studies have extracted morphological information of spectral imaging as spatial information, and combined with mean spectral information to perform modeling. The method belongs to the preliminary exploration of a space-spectrum combined model, but the space-spectrum combined model still has huge research space because morphological information belongs to lower-level image information.
Therefore, the embodiment of the invention provides an unsupervised anomaly detection method and system based on double-current joint density estimation, which adopts a one-dimensional self-coding network to extract average spectral features, adopts a two-dimensional self-coding network to extract main component image spatial features, and then performs fusion and density estimation on double-current features of a normal sample through a joint density estimation network; and designing a balance penalty loss function to balance the contribution balance of the spectral feature and the spatial feature and prevent a certain branch from being over-fitted. The method can realize unsupervised spectrum imaging data anomaly detection based on 'space spectrum information' combination. The following description and description will proceed with reference being made to various embodiments.
Fig. 1 shows an unsupervised anomaly detection method based on dual-flow joint density estimation, which can be applied to hyperspectral nondestructive detection, food adulteration detection, fruit sugar content detection, agricultural product quality detection, fake drug identification, drug component analysis and fake drug identification, microorganism content detection and organic matter content detection, and oil and gas component analysis, and includes:
s1, collecting spectral imaging data of a sample to be detected, and constructing a sample set based on the spectral imaging data;
specifically, the method comprises the following steps:
s11, collecting first spectral imaging data of a normal sample and second spectral imaging data of an abnormal sample;
if 900 strawberry samples are collected in total, 600 normal strawberry samples and 300 abnormal strawberry samples are collected; the abnormal samples specifically comprise 100 bruising samples, 100 fungal infection samples and 100 soil pollution samples; the hyperspectral band is 900nm-1700nm, 256 channels are totally provided, the head and tail 100nm high-noise bands are removed, and 180 spectral features are totally used for modeling.
Segmenting the first spectral imaging data based on a watershed algorithm to obtain effective pixels of each normal sample and each abnormal sample; and putting the strawberry effective hyperspectral image obtained by segmentation into a blank background of 120 multiplied by 120 pixel.
S12, constructing a training sample set based on the first spectral imaging data, and constructing a test sample set based on the first spectral imaging data and the second spectral imaging data; the training sample set only comprises normal samples, and the testing sample set comprises normal samples and abnormal samples; wherein 300 normal samples are randomly selected as a training set, and the remaining 300 normal samples and 300 abnormal samples are selected as a testing set.
Averaging the spectrums of all effective pixels in a training sample set and a testing sample set to obtain an average spectrum, and taking the average spectrum as the input of the one-dimensional self-coding network; and performing principal component analysis on all effective pixels in the training sample set, and performing data dimensionality reduction to obtain the first n principal component PCA images (such as the first 3) to be used as the input of the two-dimensional self-coding network.
Step S21, constructing a double-flow joint density estimation network, as shown in FIG. 2, wherein the double-flow joint density estimation network comprises a one-dimensional self-coding network, a two-dimensional self-coding network and a joint density estimation network;
the one-dimensional self-encoding network is used for encoding, decoding and reconstructing an average spectrum of spectral imaging data, and the specific structure of the one-dimensional self-encoding network is as follows: one-dimensional convolution layer-one-dimensional deconvolution layer, the step length of the one-dimensional convolution layer and the one-dimensional deconvolution layer being 2.
The two-dimensional self-coding network is used for coding, decoding and reconstructing the first n Principal component PCA (Principal component Analysis) images of spectral imaging data, and the specific structure of the two-dimensional self-coding network is as follows: two-dimensional convolution layer-two-dimensional deconvolution layer, the step size of the two-dimensional convolution layer and the two-dimensional deconvolution layer being 2 x 2.
The coding characteristic obtained by the one-dimensional self-coding network, namely the output characteristic of the third one-dimensional convolutional layer is c 1D . The coding characteristic obtained by the two-dimensional self-coding network, namely the output characteristic of the third two-dimensional convolution layer is c 2D 。c 1D And c 2D And merging in a cascading mode to serve as the input of the joint density estimation network.
The joint density estimation network is used for performing double-current feature fusion and density estimation on the average spectral coding features obtained by the one-dimensional self-coding network and the PCA spatial coding features obtained by the two-dimensional self-coding network;
the joint density estimation network comprises m 1 +m 2 A fully connected layer, wherein the first m 1 The full connection layer is a feature fusion layer 2 The full connection layer is a density estimation layer; the coding characteristics c obtained by the one-dimensional self-coding network 1D And coding characteristics c obtained from two-dimensional self-coding network 2D Cascading as the input of a feature fusion layer, and taking the obtained fusion feature c as the input of a density estimation layer; the joint density estimationThe network density estimation layer comprises 2 full-connection layers and an output layer, the number of nodes of the full-connection layers is 16, the output layer comprises K nodes corresponding to K Gaussian partial models, and the output layer adopts a softmax function as nonlinear excitation.
S22, constructing a loss function of the double-current joint density estimation network:
L＝α 1 *L AE_1D +α 2 *L AE_2D +α 3 *L GMMnet +α 4 *L P
wherein L is AE_1D 、L AE_2D 、L GMMnet 、L P Respectively a one-dimensional self-coding network loss function, a two-dimensional self-coding network loss function, a joint density estimation network loss function, a balance penalty loss function, alpha 1 、α 2 、α 3 、α 4 To take on a value of [0,1]Constant coefficient of (d) between.
Preferably, the one-dimensional self-coding network loss function L AE_1D Comprises the following steps:
Figure GDA0003991714870000081
wherein the content of the first and second substances,
Figure GDA0003991714870000082
the spectrum is averaged for the ith training sample,
Figure GDA0003991714870000083
obtaining a reconstructed spectrum for the ith training sample average spectrum after passing through a one-dimensional self-coding network;
the two-dimensional self-coding network loss function L AE_2D Comprises the following steps:
Figure GDA0003991714870000084
wherein, the first and the second end of the pipe are connected with each other,
Figure GDA0003991714870000085
is as followsThe PCA images of the i training samples,
Figure GDA0003991714870000086
obtaining a reconstructed PCA image for the first n principal component PCA images of the ith training sample after passing through a two-dimensional self-coding network;
the joint density estimation network loss function L GMMnet Comprises the following steps:
Figure GDA0003991714870000087
wherein, c i The average spectral coding characteristics of the ith sample and the PCA spatial coding characteristics are fused to form a combined coding characteristic;
Figure GDA0003991714870000088
Σ k respectively predicting probability, mean value and covariance of the kth sub-model in the multivariate mixture Gaussian model; the specific calculation method comprises the following steps:
Figure GDA0003991714870000089
Figure GDA00039917148700000810
Figure GDA0003991714870000091
the equilibrium penalty loss function L P Comprises the following steps:
Figure GDA0003991714870000092
wherein, w 1D,l Estimating c in network feature fusion layer for joint density 1D Corresponding network weight, w 2D,s Estimating c in network feature fusion layer for joint density 2D The corresponding network weight. Equalizing penalty loss function L p The existing method aims to balance the contributions of two feature extraction branches and avoid the phenomena that the weight of a certain branch is too large and the fusion is invalid.
S3, training the double-current joint density estimation network based on the sample set; training a double-current joint density estimation network by using normal samples in a training sample set by adopting a gradient descent method, and estimating and optimizing a one-dimensional self-coding network, a two-dimensional self-coding network and the joint density network in the training process; training 2000 epochs in total, with a learning rate of 1 × 10 -3 ；
And S4, carrying out sample prediction based on the trained double-current joint density estimation network.
The step S4 specifically includes:
inputting the prediction sample into a trained double-current joint density estimation network, performing one-dimensional coding and two-dimensional coding on the prediction sample, and performing joint density estimation network to obtain a likelihood function E (c) of the prediction sample i )；
Figure GDA0003991714870000093
Wherein, c t Obtaining joint coding characteristics for a prediction sample through a one-dimensional self-coding network and a two-dimensional self-coding network;
Figure GDA0003991714870000094
outputting a predicted value of a k Gaussian component model of a prediction sample, namely the k node output of the double-current joint density estimation network;
Figure GDA0003991714870000095
∑ k,prior respectively is the prior mean value and the prior covariance of the kth Gaussian mixture model in the multi-element mixed Gaussian model obtained by training.
And taking the likelihood function as an abnormal score, if the abnormal score is larger than a threshold th, judging that the prediction sample is abnormal, otherwise, judging that the prediction sample is normal.
The value method of th in this embodiment is: hypothesis training set sample likelihood function E (c) i ) And (3) according with gamma distribution, estimating the gamma distribution parameters by utilizing a training set fruit, and determining that the threshold th of abnormal detection is 95% of the gamma distribution.
And (4) carrying out 10 times of random sampling and corresponding training on the normal sample, and averaging the abnormal detection results for model evaluation. A one-dimensional self-coding network (AE _ 1D) and a two-dimensional self-coding network (AE _ 2D) are used as two comparison methods. In both comparison methods, the reconstruction error is used as an abnormal score, and the abnormal score threshold value-taking method is consistent with the method. The model was evaluated using area under the curve (AUC), precision (Precision) and Recall (Recall). The results of 10 modeling calculations are compared in table 1.
Method AUC Precision RECALL
AE_1D 75.4％ 71.6％ 67.3％
AE_2D 73.1％ 70.7％ 66.0％
Method of an embodiment of the invention 86.2％ 77.7％ 80.3％
The calculation result shows that the average AUC of the data set obtained by the method provided by the embodiment of the invention is 75.4%, the average AUC obtained by the one-dimensional self-coding network is 73.1%, the average AUC obtained by the two-dimensional self-coding network is 86.2%, and the accuracy of anomaly identification is remarkably improved.
The embodiment of the invention also provides an unsupervised anomaly detection system based on double-stream joint density estimation, and the unsupervised anomaly detection method based on double-stream joint density estimation in each embodiment comprises the following steps:
the sample acquisition module is used for acquiring spectral imaging data of a sample to be detected and constructing a sample set based on the spectral imaging data;
the network construction module is used for constructing a double-flow joint density estimation network, and the double-flow joint density estimation network comprises a one-dimensional self-coding network, a two-dimensional self-coding network and a joint density estimation network; the one-dimensional self-coding network is used for coding, decoding and reconstructing an average spectrum of spectral imaging data, the two-dimensional self-coding network is used for coding, decoding and reconstructing the first n principal component PCA images of the spectral imaging data, and the joint density estimation network is used for performing double-current feature fusion and density estimation on average spectral coding features obtained by the one-dimensional self-coding network and PCA space coding features obtained by the two-dimensional self-coding network;
the training module is used for training the double-current joint density estimation network based on the sample set;
and the prediction module is used for predicting samples based on the trained double-flow joint density estimation network.
Based on the same concept, an embodiment of the present invention further provides an entity structure schematic diagram, as shown in fig. 3, the server may include: a processor (processor) 810, a communication Interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication Interface 820 and the memory 830 communicate with each other via the communication bus 840. The processor 810 may invoke logic instructions in the memory 830 to perform the steps of the unsupervised anomaly detection method based on dual stream joint density estimation as described in the various embodiments above. Examples include:
s1, collecting spectral imaging data of a sample to be detected, and constructing a sample set based on the spectral imaging data;
s2, constructing a double-flow joint density estimation network, wherein the double-flow joint density estimation network comprises a one-dimensional self-coding network, a two-dimensional self-coding network and a joint density estimation network; the one-dimensional self-coding network is used for coding, decoding and reconstructing an average spectrum of spectral imaging data, the two-dimensional self-coding network is used for coding, decoding and reconstructing the first n principal component PCA images of the spectral imaging data, and the joint density estimation network is used for performing double-current feature fusion and density estimation on average spectral coding features obtained by the one-dimensional self-coding network and PCA space coding features obtained by the two-dimensional self-coding network;
s3, training the double-current joint density estimation network based on the sample set;
and S4, carrying out sample prediction based on the trained double-current joint density estimation network.
In addition, the logic instructions in the memory 830 can be implemented in the form of software functional units and stored in a computer readable storage medium when the software functional units are sold or used as independent products. Based on such understanding, the technical solution of the present invention or a part thereof which substantially contributes to the prior art may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, and various media capable of storing program codes.
Based on the same concept, embodiments of the present invention further provide a non-transitory computer-readable storage medium storing a computer program, where the computer program includes at least one code, where the at least one code is executable by a master device to control the master device to implement the steps of the unsupervised anomaly detection method based on dual-stream joint density estimation according to the foregoing embodiments. Examples include:
s1, collecting spectral imaging data of a sample to be detected, and constructing a sample set based on the spectral imaging data;
s2, constructing a double-flow joint density estimation network, wherein the double-flow joint density estimation network comprises a one-dimensional self-coding network, a two-dimensional self-coding network and a joint density estimation network; the one-dimensional self-coding network is used for coding, decoding and reconstructing an average spectrum of spectral imaging data, the two-dimensional self-coding network is used for coding, decoding and reconstructing the first n principal component PCA images of the spectral imaging data, and the joint density estimation network is used for performing double-current feature fusion and density estimation on average spectral coding features obtained by the one-dimensional self-coding network and PCA space coding features obtained by the two-dimensional self-coding network;
s3, training the double-current joint density estimation network based on the sample set;
and S4, carrying out sample prediction based on the trained double-current joint density estimation network.
Based on the same technical concept, the embodiment of the present application further provides a computer program, which is used to implement the above method embodiment when the computer program is executed by the main control device.
The program may be stored in whole or in part on a storage medium packaged with the processor, or in part or in whole on a memory not packaged with the processor.
Based on the same technical concept, the embodiment of the present application further provides a processor, and the processor is configured to implement the above method embodiment. The processor may be a chip.
In summary, according to the unsupervised anomaly detection method and system based on the double-current joint density estimation provided by the embodiment of the invention, the overall average spectrum and the plurality of block average spectra share the same neural network weight, and the mean square error of the overall average spectrum and the mean square error of the block average predicted value are reduced simultaneously in the training process; a block smooth loss function is designed, and the addition of the loss function can improve the continuity and smoothness of block prediction, inhibit the sudden change of the predicted values of adjacent blocks and enable the predicted values of the adjacent blocks to be in smooth transition; by utilizing the prior spatial information, the anti-noise capability of the network can be further improved, and the prediction precision and the model robustness are improved.
The embodiments of the present invention can be arbitrarily combined to achieve different technical effects.
In the above embodiments, all or part of the implementation may be realized by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the procedures or functions described in accordance with the present application are generated, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (e.g., coaxial cable, fiber optic, digital subscriber line) or wirelessly (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that incorporates one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk), among others.
One of ordinary skill in the art will appreciate that all or part of the processes in the methods of the above embodiments may be implemented by hardware related to instructions of a computer program, which may be stored in a computer-readable storage medium, and when executed, may include the processes of the above method embodiments. And the aforementioned storage medium includes: various media capable of storing program codes, such as ROM or RAM, magnetic or optical disks, etc.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (7)

1. An unsupervised anomaly detection method based on double-flow joint density estimation is characterized by comprising the following steps:
s1, collecting spectral imaging data of a sample to be detected, and constructing a sample set based on the spectral imaging data;
s2, constructing a double-flow joint density estimation network, wherein the double-flow joint density estimation network comprises a one-dimensional self-coding network, a two-dimensional self-coding network and a joint density estimation network; the one-dimensional self-coding network is used for coding, decoding and reconstructing an average spectrum of spectral imaging data, the two-dimensional self-coding network is used for coding, decoding and reconstructing the first n principal component PCA images of the spectral imaging data, and the joint density estimation network is used for performing double-current feature fusion and density estimation on average spectral coding features obtained by the one-dimensional self-coding network and PCA space coding features obtained by the two-dimensional self-coding network;
constructing a loss function of the double-flow joint density estimation network:
L＝α 1 *L AE_1D +α 2 *L AE_2D +α 3 *L GMMnet +α 4 *L P
wherein L is AE_1D 、L AE_2D 、L GMMnet 、L P Respectively a one-dimensional self-coding network loss function, a two-dimensional self-coding network loss function, a joint density estimation network loss function, a balance penalty loss function, alpha 1 、α 2 、α 3 、α 4 To take a value of [0,1]Constant coefficient of between;
the joint density estimation network comprises m 1 +m 2 A fully connected layer, wherein the first m 1 The full connection layer is a feature fusion layer 2 Each full connection layer is a density estimation layer; coding characteristics c obtained by the one-dimensional self-coding network 1D And coding characteristics c obtained from two-dimensional self-coding network 2D Cascading is carried out and used as the input of a characteristic fusion layer, and the obtained fusion characteristic c is used as the input of a density estimation layer; the output layer of the joint density estimation network comprises K nodes corresponding to K Gaussian partial models, and the output layer adopts a softmax function as nonlinear excitation;
the one-dimensional self-coding network loss function L AE_1D Comprises the following steps:
Figure FDA0003991714860000011
wherein, the first and the second end of the pipe are connected with each other,
Figure FDA0003991714860000012
the spectrum is averaged for the ith training sample,
Figure FDA0003991714860000013
passing the ith training sample average spectrum through a one-dimensional self-coding network to obtain a reconstructed spectrum;
the two-dimensional self-coding network loss function L AE_2D Comprises the following steps:
Figure FDA0003991714860000014
wherein the content of the first and second substances,
Figure FDA0003991714860000015
for the PCA image of the ith training sample,
Figure FDA0003991714860000016
carrying out two-dimensional self-coding on the first n principal component PCA images of the ith training sample to obtain a reconstructed PCA image;
the joint density estimation network loss function L GMMnet Comprises the following steps:
Figure FDA0003991714860000021
wherein, c i The average spectral coding characteristics of the ith sample and the PCA spatial coding characteristics are fused to form combined coding characteristics;
Figure FDA0003991714860000022
Σ k respectively predicting probability, mean value and covariance of the kth sub-model in the multivariate mixture Gaussian model; the specific calculation method comprises the following steps:
Figure FDA0003991714860000023
Figure FDA0003991714860000024
Figure FDA0003991714860000025
in the above formula, the first and second carbon atoms are,
Figure FDA0003991714860000026
a predicted value of a k Gaussian component model of a training sample is obtained; the balance penalty loss function L P Comprises the following steps:
Figure FDA0003991714860000027
wherein, w 1D,l Estimating c in network feature fusion layer for joint density 1D Corresponding network weight, w 2D,s Estimating c in network feature fusion layer for joint density 2D The corresponding network weight;
s3, training the double-current joint density estimation network based on the sample set;
and S4, predicting a sample based on the trained double-flow joint density estimation network.
2. The unsupervised anomaly detection method based on dual-stream joint density estimation according to claim 1, wherein the step S1 specifically comprises:
acquiring spectral imaging data of a normal sample to construct a first spectral imaging data set, and acquiring spectral imaging data of an abnormal sample to construct a second spectral imaging data set;
based on a watershed algorithm, all samples in the first spectral imaging data set and the second spectral imaging data set are segmented to obtain effective pixels of each normal sample and each abnormal sample;
constructing a training sample set based on a part of samples in the first spectral imaging data set, and constructing a test sample set based on the rest of samples in the first spectral imaging data set and the second spectral imaging data;
averaging the spectrums of all effective pixels in a training sample set and a testing sample set to obtain an average spectrum, and taking the average spectrum as the input of the one-dimensional self-coding network; and performing principal component analysis on all effective pixels in the training sample set, and performing data dimension reduction to obtain the first n principal component PCA images as the input of the two-dimensional self-coding network.
3. The unsupervised anomaly detection method based on dual-stream joint density estimation according to claim 1, wherein the step S4 specifically comprises:
inputting the prediction sample into a trained double-flow joint density estimation network to obtain a likelihood function of the prediction sample;
and taking the likelihood function as an abnormal score, if the abnormal score is larger than a threshold th, judging that the prediction sample is abnormal, otherwise, judging that the prediction sample is normal.
4. The unsupervised anomaly detection method based on dual-stream joint density estimation according to claim 3, wherein in step S4, a likelihood function E (c) of samples is predicted t ) Comprises the following steps:
Figure FDA0003991714860000031
wherein, c i The combined coding feature after the average spectral coding feature and the PCA spatial coding feature of the ith sample are fused, c t Obtaining joint coding characteristics for a prediction sample through a one-dimensional self-coding network and a two-dimensional self-coding network;
Figure FDA0003991714860000032
outputting a predicted value of a k Gaussian branch model of a prediction sample, namely the k node output of the double-flow joint density estimation network;
Figure FDA0003991714860000033
∑ k,prior respectively the prior mean and the prior covariance of the kth Gaussian mixture model in the multi-element mixed Gaussian model obtained by training.
5. An unsupervised anomaly detection system based on dual-stream joint density estimation, comprising:
the sample acquisition module is used for acquiring spectral imaging data of a sample to be detected and constructing a sample set based on the spectral imaging data;
the network construction module is used for constructing a double-flow joint density estimation network, and the double-flow joint density estimation network comprises a one-dimensional self-coding network, a two-dimensional self-coding network and a joint density estimation network; the one-dimensional self-coding network is used for coding, decoding and reconstructing an average spectrum of spectral imaging data, the two-dimensional self-coding network is used for coding, decoding and reconstructing the first n principal component PCA images of the spectral imaging data, and the joint density estimation network is used for performing double-current feature fusion and density estimation on average spectral coding features obtained by the one-dimensional self-coding network and PCA space coding features obtained by the two-dimensional self-coding network;
constructing a loss function of the double-flow joint density estimation network:
L＝α 1 *L AE_1D +α 2 *L AE_2D +α 3 *L GMMnet +α 4 *L P
wherein L is AE_1D 、L AE_2D 、L GMMnet 、L P Respectively a one-dimensional self-coding network loss function, a two-dimensional self-coding network loss function, a joint density estimation network loss function, a balance penalty loss function, alpha 1 、α 2 、α 3 、α 4 To take on a value of [0,1]Constant coefficient of between;
the joint density estimation network comprises m 1 +m 2 A fully connected layer, wherein the first m 1 The full connection layer is a feature fusion layer 2 Each full connection layer is a density estimation layer; coding characteristics c obtained by the one-dimensional self-coding network 1D And coding characteristics c obtained from two-dimensional self-coding network 2D Cascading is carried out and used as the input of a characteristic fusion layer, and the obtained fusion characteristic c is used as the input of a density estimation layer; of said joint density estimation networkThe output layer comprises K nodes corresponding to K Gaussian partial models, and the output layer adopts a softmax function as nonlinear excitation;
the one-dimensional self-coding network loss function L AE_1D Comprises the following steps:
Figure FDA0003991714860000041
wherein, the first and the second end of the pipe are connected with each other,
Figure FDA0003991714860000042
the spectrum is averaged for the ith training sample,
Figure FDA0003991714860000043
passing the ith training sample average spectrum through a one-dimensional self-coding network to obtain a reconstructed spectrum;
the two-dimensional self-coding network loss function L AE_2D Comprises the following steps:
Figure FDA0003991714860000044
wherein the content of the first and second substances,
Figure FDA0003991714860000045
for the PCA image of the ith training sample,
Figure FDA0003991714860000046
obtaining a reconstructed PCA image for the first n principal component PCA images of the ith training sample after passing through a two-dimensional self-coding network;
the joint density estimation network loss function L GMMnet Comprises the following steps:
Figure FDA0003991714860000047
wherein, c i Encoding the average spectrum of the ith sample with codeCombining the PCA spatial coding features and the combined coding features after the PCA spatial coding features are fused;
Figure FDA0003991714860000048
Σ k respectively predicting probability, mean value and covariance of the kth sub-model in the multivariate mixture Gaussian model; the specific calculation method comprises the following steps:
Figure FDA0003991714860000049
Figure FDA00039917148600000410
Figure FDA0003991714860000051
in the above formula, the first and second carbon atoms are,
Figure FDA0003991714860000052
a predicted value of a k Gaussian component model of a training sample is obtained; the equilibrium penalty loss function L P Comprises the following steps:
Figure FDA0003991714860000053
wherein, w 1D,l Estimating c in network feature fusion layer for joint density 1D Corresponding network weight, w 2D,s Estimating c in network feature fusion layer for joint density 2D The corresponding network weight;
the training module is used for training the double-current joint density estimation network based on the sample set;
and the prediction module is used for predicting samples based on the trained double-flow joint density estimation network.
6. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program implements the steps of the method for unsupervised anomaly detection based on dual-stream joint density estimation according to any of claims 1 to 4.
7. A non-transitory computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the unsupervised anomaly detection method based on dual-stream joint density estimation according to any of claims 1 to 4.
CN202110817301.3A 2021-07-20 2021-07-20 Unsupervised anomaly detection method and system based on double-flow joint density estimation Active CN113762333B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110817301.3A CN113762333B (en) 2021-07-20 2021-07-20 Unsupervised anomaly detection method and system based on double-flow joint density estimation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110817301.3A CN113762333B (en) 2021-07-20 2021-07-20 Unsupervised anomaly detection method and system based on double-flow joint density estimation

Publications (2)

Publication Number Publication Date
CN113762333A CN113762333A (en) 2021-12-07
CN113762333B true CN113762333B (en) 2023-02-28

Family

ID=78787682

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110817301.3A Active CN113762333B (en) 2021-07-20 2021-07-20 Unsupervised anomaly detection method and system based on double-flow joint density estimation

Country Status (1)

Country Link
CN (1) CN113762333B (en)

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109949278A (en) * 2019-03-06 2019-06-28 西安电子科技大学 Hyperspectral abnormity detection method based on confrontation autoencoder network
CN110298235A (en) * 2019-05-17 2019-10-01 中国科学院西安光学精密机械研究所 Hyperspectral abnormity detection method and system based on manifold constraint autoencoder network
WO2020031570A1 (en) * 2018-08-10 2020-02-13 日本電信電話株式会社 Abnormality detection device, probability distribution learning device, self-encoder learning device, data conversion device, and program
EP3620983A1 (en) * 2018-09-05 2020-03-11 Sartorius Stedim Data Analytics AB Computer-implemented method, computer program product and system for data analysis
CN111127331A (en) * 2019-10-22 2020-05-08 广东启迪图卫科技股份有限公司 Image denoising method based on pixel-level global noise estimation coding and decoding network
CN111179235A (en) * 2019-12-23 2020-05-19 沈阳先进医疗设备技术孵化中心有限公司 Image detection model generation method and device, and application method and device
CN111539447A (en) * 2020-03-17 2020-08-14 广东省智能制造研究所 Hyperspectrum and terahertz data depth fusion-based classification method
CN111626317A (en) * 2019-08-14 2020-09-04 广东省智能制造研究所 Semi-supervised hyperspectral data analysis method based on double-flow conditional countermeasure generation network
CN112183561A (en) * 2020-11-09 2021-01-05 山东中医药大学 Joint fusion subtraction automatic encoder algorithm for image feature extraction
CN112329685A (en) * 2020-11-16 2021-02-05 常州大学 Method for detecting crowd abnormal behaviors through fusion type convolutional neural network
CN112924749A (en) * 2021-02-04 2021-06-08 西安电子科技大学 Unsupervised counterstudy electromagnetic spectrum abnormal signal detection method

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107180248A (en) * 2017-06-12 2017-09-19 桂林电子科技大学 Strengthen the hyperspectral image classification method of network based on associated losses
US10999247B2 (en) * 2017-10-24 2021-05-04 Nec Corporation Density estimation network for unsupervised anomaly detection
US10373056B1 (en) * 2018-01-25 2019-08-06 SparkCognition, Inc. Unsupervised model building for clustering and anomaly detection
CN112381756B (en) * 2020-09-30 2022-12-16 广东省科学院智能制造研究所 Hyperspectral data analysis method and system based on block smoothing neural network

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020031570A1 (en) * 2018-08-10 2020-02-13 日本電信電話株式会社 Abnormality detection device, probability distribution learning device, self-encoder learning device, data conversion device, and program
EP3620983A1 (en) * 2018-09-05 2020-03-11 Sartorius Stedim Data Analytics AB Computer-implemented method, computer program product and system for data analysis
CN109949278A (en) * 2019-03-06 2019-06-28 西安电子科技大学 Hyperspectral abnormity detection method based on confrontation autoencoder network
CN110298235A (en) * 2019-05-17 2019-10-01 中国科学院西安光学精密机械研究所 Hyperspectral abnormity detection method and system based on manifold constraint autoencoder network
CN111626317A (en) * 2019-08-14 2020-09-04 广东省智能制造研究所 Semi-supervised hyperspectral data analysis method based on double-flow conditional countermeasure generation network
CN111127331A (en) * 2019-10-22 2020-05-08 广东启迪图卫科技股份有限公司 Image denoising method based on pixel-level global noise estimation coding and decoding network
CN111179235A (en) * 2019-12-23 2020-05-19 沈阳先进医疗设备技术孵化中心有限公司 Image detection model generation method and device, and application method and device
CN111539447A (en) * 2020-03-17 2020-08-14 广东省智能制造研究所 Hyperspectrum and terahertz data depth fusion-based classification method
CN112183561A (en) * 2020-11-09 2021-01-05 山东中医药大学 Joint fusion subtraction automatic encoder algorithm for image feature extraction
CN112329685A (en) * 2020-11-16 2021-02-05 常州大学 Method for detecting crowd abnormal behaviors through fusion type convolutional neural network
CN112924749A (en) * 2021-02-04 2021-06-08 西安电子科技大学 Unsupervised counterstudy electromagnetic spectrum abnormal signal detection method

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
A Hybrid Autoencoder and Density Estimation Model for Anomaly Detection;Van Loi Cao et al;《International Conference on Parallel Problem Sloving from Nature》;20160831;717-726 *
E2E-LIADE: End-to-End Local Invariant Autoencoding Density Estimation Model for Anomaly Target Detection in Hyperspectral Image;Kai Jiang et al;《IEEE Transactions on Cybernetics ( Early Access )》;20210602;1-12 *
基于多尺度空谱融合网络的高光谱图像分类;梁苗苗;《中国博士学位论文全文数据库 工程科技Ⅱ辑》;20190715;第2019年卷(第7期);C028-7 *
基于高光谱空-谱联合信息的舰船检测算法研究;代嘉慧;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;20210715;第2021年卷(第7期);C028-79 *
深度随机神经网络单分类器理论、算法与应用研究;戴浩桢;《中国优秀硕士学位论文全文数据库 信息科技辑》;20210215;第2021年卷(第2期);I140-172 *

Also Published As

Publication number Publication date
CN113762333A (en) 2021-12-07

Similar Documents

Publication Publication Date Title
Jiang et al. A fully convolutional neural network-based regression approach for effective chemical composition analysis using near-infrared spectroscopy in cloud
WO2022268102A1 (en) Deep learning-based cancer prognosis survival prediction method and device, and storage medium
CN112381756B (en) Hyperspectral data analysis method and system based on block smoothing neural network
CN111274865B (en) Remote sensing image cloud detection method and device based on full convolution neural network
CN111626317B (en) Semi-supervised hyperspectral data analysis method based on double-flow conditional countermeasure generation network
CN111667489A (en) Cancer hyperspectral image segmentation method and system based on double-branch attention deep learning
CN111161362B (en) Spectral image identification method for growth state of tea tree
CN113435253B (en) Multi-source image combined urban area ground surface coverage classification method
CN111460956A (en) Unbalanced electrocardiogram sample classification method based on data enhancement and loss weighting
Jie et al. Research on Citrus grandis granulation determination based on hyperspectral imaging through deep learning
Abdolmaleki et al. Selecting optimum base wavelet for extracting spectral alteration features associated with porphyry copper mineralization using hyperspectral images
CN113030001B (en) Fruit sugar degree detection method and system
CN113095409A (en) Hyperspectral image classification method based on attention mechanism and weight sharing
CN114612715A (en) Edge federal image classification method based on local differential privacy
CN114494846A (en) Hyperspectral image classification method based on multi-scale internal volume network
CN113762333B (en) Unsupervised anomaly detection method and system based on double-flow joint density estimation
CN113591816B (en) Hyperspectral anomaly detection method and system based on self-supervision guide coding network
CN112418256A (en) Classification, model training and information searching method, system and equipment
Tsakiridis et al. In situ grape ripeness estimation via hyperspectral imaging and deep autoencoders
CN116825363B (en) Early lung adenocarcinoma pathological type prediction system based on fusion deep learning network
CN111539447B (en) Hyperspectrum and terahertz data depth fusion-based classification method
CN109460788B (en) Hyperspectral image classification method based on low-rank-sparse information combination network
CN112801187B (en) Hyperspectral data analysis method and system based on attention mechanism and ensemble learning
CN111968087A (en) Plant disease area detection method
CN115019215A (en) Hyperspectral image-based soybean disease and pest identification method and device

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