CN117574262A - Underwater sound signal classification method, system and medium for small sample problem - Google Patents

Underwater sound signal classification method, system and medium for small sample problem Download PDF

Info

Publication number
CN117574262A
CN117574262A CN202311406647.XA CN202311406647A CN117574262A CN 117574262 A CN117574262 A CN 117574262A CN 202311406647 A CN202311406647 A CN 202311406647A CN 117574262 A CN117574262 A CN 117574262A
Authority
CN
China
Prior art keywords
sample
sample set
classifier
sound signal
samples
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202311406647.XA
Other languages
Chinese (zh)
Inventor
邹丹
张卫华
朱敏
陆丽娜
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.)
National University of Defense Technology
Original Assignee
National University of Defense Technology
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 National University of Defense Technology filed Critical National University of Defense Technology
Priority to CN202311406647.XA priority Critical patent/CN117574262A/en
Publication of CN117574262A publication Critical patent/CN117574262A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—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
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/30—Assessment of water resources

Landscapes

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

Abstract

The invention discloses a small sample problem-oriented underwater acoustic signal classification method, a system and a medium, wherein the method comprises the steps of randomly sampling a marked underwater acoustic signal sample set and training a classifier C s The method comprises the steps of carrying out a first treatment on the surface of the Using classifier C s Carrying out classification probability prediction and sample migration on unlabeled sample sets, training an initial classifier of the constructed positive and negative sample sets, and calculating index values; and carrying out classification probability prediction on the current positive and negative sample set by using the i-1 th classifier, carrying out sample migration, training the i-th classifier based on the newly constructed positive and negative sample set, calculating an index value, and selecting whether to continue iteration according to the index value. The invention aims to convert the problem of small samples into the problem of traditional supervision and semi-supervision learning, and dig reliable normal underwater sound from unlabeled sample set by means of labeled samplesThe scale of the labeling sample set for training the machine learning model is expanded by the signal sample and the abnormal underwater sound signal sample, so that the classification precision of the abnormal underwater sound signal sample is improved.

Description

Underwater sound signal classification method, system and medium for small sample problem
Technical Field
The invention relates to the technical field of underwater acoustic signal detection, in particular to an underwater acoustic signal classification method, system and medium for small sample problem.
Background
The classification of underwater acoustic signals is a key technique for achieving underwater acoustic target recognition. In a typical application scene of abnormal underwater acoustic signal detection, because the generation environment of underwater acoustic signal data is complex, the data updating speed is high, the data marking difficulty is high, and a high-quality large-scale labeling sample data set cannot be established for each specific abnormal underwater acoustic signal detection learning task, only a small number of labeling samples can be used for training a machine learning model in actual application, and the generalization capability of the model is greatly reduced. How to use a limited marked sample and a large number of unmarked samples to improve the detection precision of the abnormal underwater sound signal is one of the key problems to be solved in the field of abnormal underwater sound signal detection.
Disclosure of Invention
The invention aims to solve the technical problems: aiming at the problems in the prior art, the invention provides a small sample problem-oriented underwater acoustic signal classification method, a small sample problem-oriented underwater acoustic signal classification system and a medium, and aims to convert the small sample problem into the traditional supervised and semi-supervised learning problem, dig reliable normal underwater acoustic signal samples and abnormal underwater acoustic signal samples from unlabeled sample sets by means of labeled samples, expand the scale of the labeled sample set for training a machine learning model, and further improve the classification precision of the abnormal underwater acoustic signal samples.
In order to solve the technical problems, the invention adopts the following technical scheme:
a method for classifying underwater acoustic signals for small sample problems, comprising:
s1, randomly sampling a marked underwater sound signal sample set L;
s2, training classifier C based on newly constructed positive and negative sample sets of random sampling s ；
S3, using classifier C s Carrying out classification probability prediction on the unlabeled sample set U and carrying out sample migration;
s4, training an initial classifier C based on newly constructed positive and negative sample sets of sample migration 0 And calculates an initial classifier C 0 Corresponding index value F 0 The method comprises the steps of carrying out a first treatment on the surface of the Initializing the value of an iteration counter i;
s5, using the i-1 th classifier C i-1 Carrying out classification probability prediction on the current positive and negative sample sets and carrying out sample migration;
s6, training an ith classifier C based on newly constructed positive and negative sample sets of sample migration i And calculates the ith classifier C i Corresponding index value F i ；
S7, if the ith classifier C i Index value F of (2) i Greater than the i-1 th classifier C i-1 Index value F of (2) i-1 The iteration counter i is increased by 1, and the step S5 is continued to iterate; otherwise, the i-1 th classifier C i-1 As an optimal classifier for classifying unlabeled underwater acoustic signals.
Optionally, step S1 includes: dividing the marked underwater sound signal sample set L into an abnormal underwater sound signal sample set L P And a set L of normal underwater sound signal samples N In a given quantitative proportion r p From set L P Randomly selected part of the samples form a set S P In a given quantitative proportion r n From set L N Randomly selected part of the samples form a set S N 。
Optionally, step S2 includes: using L P -S P As a positive sample set and given tag 1, u+s is used P +S N As a negative sample set and given a label 0, training the positive sample set and the negative sample set by using a supervised learning method to obtain a classifier C s Wherein U is an unlabeled sample set, and the unlabeled sample set contains unlabeled underwater sound signal samples.
Optionally, step S3 includes: for set S P All samples in (1) use classifier C s Predicting the probability of the abnormal underwater sound signal sample, and recording the maximum abnormal probability value t p,max And a minimum anomaly probability value t p,min The method comprises the steps of carrying out a first treatment on the surface of the For all samples in unlabeled sample set U, classifier C is used s For which it is an abnormal underwater sound signal sampleThe probability of the cost is predicted and combined with the maximum abnormal probability value t p,max And a minimum anomaly probability value t p,min Migrating samples in unlabeled sample set U to set L P And L N In (a) and (b); for set S N All samples in (1) use classifier C s Predicting the probability of the signal sample being a normal underwater sound signal sample, and recording the maximum normal probability value t n,max And a minimum normal probability value t n,min The method comprises the steps of carrying out a first treatment on the surface of the For all samples in unlabeled sample set U, classifier C is used s Predicting the probability of the signal sample being a normal underwater sound signal sample, and combining the maximum normal probability value t n,max And a minimum normal probability value t n,min Migrating samples in unlabeled sample set U to set L P And L N Is a kind of medium.
Optionally, the combined maximum anomaly probability value t p,max And a minimum anomaly probability value t p,min Migrating samples in unlabeled sample set U to abnormal underwater sound signal sample set L P And a normal underwater sound signal sample set L N Comprises the following steps: for all s in unlabeled sample set U p1 The probability value is greater than or equal to the maximum abnormal probability value t p,max Will be s above p1 The individual samples are removed from the unlabeled sample set U and added to the abnormal underwater sound signal sample set L P The method comprises the steps of carrying out a first treatment on the surface of the For all s in unlabeled sample set U p2 The probability value is smaller than the maximum abnormal probability value t p,max And is greater than the minimum abnormal probability value t p,min If s is satisfied p2 ≤c p *s p1 Then s is as described above p2 The individual samples are removed from the unlabeled sample set U and added to the normal underwater sound signal sample set L N If s p2 >c p *s p1 Then will first go above s p2 Sorting the samples from big to small according to probability value, and then sorting the previous c p *s p1 The individual samples are removed from the unlabeled sample set U and added to the abnormal underwater sound signal sample set L P The method comprises the steps of carrying out a first treatment on the surface of the Wherein c p Is a constant parameter.
Optionally, the combined maximum normal probability value t n,max And a minimum normal probability value t n,min Migrating samples in unlabeled sample set UMove to the abnormal underwater sound signal sample set L P And a normal underwater sound signal sample set L N Comprises the following steps: for all s in unlabeled sample set U n1 The probability value is greater than or equal to the maximum normal probability value t n,max Will be s above n1 The individual samples are removed from the unlabeled sample set U and added to the normal underwater sound signal sample set L N The method comprises the steps of carrying out a first treatment on the surface of the For all s in unlabeled sample set U n2 The probability values are smaller than the maximum normal probability value t n,max And is greater than the minimum normal probability value t n,min If s is satisfied n2 ≤c n *s n1 Then s is as described above n2 The individual samples are removed from the unlabeled sample set U and added to the normal underwater sound signal sample set L N If s n2 >c n *s n1 Then will first go above s n2 Sorting the samples from big to small according to probability value, and then sorting the previous c n *s n1 The individual samples are removed from the unlabeled sample set U and added to the normal underwater sound signal sample set L N The method comprises the steps of carrying out a first treatment on the surface of the Wherein c n Is a constant parameter.
Optionally, the initial classifier C is trained based on the newly constructed positive and negative sample sets of sample migration in step S4 0 Comprising the following steps: using set L P As a positive sample set and given tag 1, set L is used N As a negative sample set and given a label 0, training by using a supervised learning method by adopting the positive sample set and the negative sample set to obtain an initial classifier C 0 。
Optionally, step S4 further comprises establishing a set L PA And L NA The two incremental training sets are initially empty, and step S5 includes: for set L P Using the i-1 th classifier C i-1 Predicting the probability of the abnormal underwater sound signal sample, and recording the minimum probability value u p The method comprises the steps of carrying out a first treatment on the surface of the For all samples in unlabeled sample set U, use the i-1 st classifier C i-1 Predicting the probability of the abnormal underwater sound signal sample, and making the probability value larger than the minimum probability value u p Samples removed from unlabeled sample set U and added to set L pA The method comprises the steps of carrying out a first treatment on the surface of the For set L N Using the i-1 th classifier C i-1 Predicting the probability of the signal sample being a normal underwater sound signal sample, and recording the minimum probability value u n The method comprises the steps of carrying out a first treatment on the surface of the For all samples in unlabeled sample set U, use the i-1 st classifier C i-1 Predicting the probability of the normal underwater sound signal sample, and making the probability value larger than the minimum probability value u n Samples are removed from set U and added to set L NA The method comprises the steps of carrying out a first treatment on the surface of the The step S6 comprises the following steps: using L P +L PA As a positive sample set and given tag 1, use L N +L NA As a negative sample set and given a label 0, training by using a supervised learning method by adopting the positive sample set and the negative sample set to obtain an ith classifier C i The method comprises the steps of carrying out a first treatment on the surface of the At set L P And L N The ith classifier C obtained by the previous calculation of the iteration of the round i Index value F of (2) i The method comprises the steps of carrying out a first treatment on the surface of the When the iteration counter i is incremented by 1 in step S7, the method further comprises the step of collecting L PA Join set L P Will be set L NA Join set L N 。
In addition, the invention also provides a small sample problem-oriented underwater sound signal classification system which comprises a microprocessor and a memory which are connected with each other, wherein the microprocessor is programmed or configured to execute the small sample problem-oriented underwater sound signal classification method.
Furthermore, the present invention provides a computer readable storage medium having stored therein a computer program for being programmed or configured by a microprocessor to perform the small sample problem oriented underwater sound signal classification method.
Compared with the prior art, the invention has the following advantages: the method comprises randomly sampling a marked underwater sound signal sample set and training a classifier C s The method comprises the steps of carrying out a first treatment on the surface of the Using classifier C s Carrying out classification probability prediction and sample migration on unlabeled sample sets, training an initial classifier of the constructed positive and negative sample sets, and calculating index values; and carrying out classification probability prediction on the current positive and negative sample set by using the i-1 th classifier, carrying out sample migration, training the i-th classifier based on the newly constructed positive and negative sample set, calculating an index value, and selecting whether to continue iteration according to the index value. The inventionThe problem of small samples is converted into the traditional supervised and semi-supervised learning problem, reliable normal underwater sound signal samples and abnormal underwater sound signal samples are mined from unlabeled sample sets by means of marked samples, the scale of the marked sample sets for training a machine learning model is expanded, and therefore classification accuracy of the abnormal underwater sound signal samples is improved.
Drawings
FIG. 1 is a schematic diagram of a basic flow of a method according to an embodiment of the present invention.
Detailed Description
Hereinafter, embodiments of the technical solution of the present invention will be described in detail by practical examples. In the practical example, the original data set contains 179 pieces of underwater sound target data (abnormal underwater sound signals) and 832 pieces of environmental noise data (normal underwater sound signals), the sampling rate is 16k, in order to facilitate the processing of the preprocessing stage, the time domain signal sampling rate of the underwater sound signals is reduced to 8k, then the frame dividing frame length is 4 seconds, the frames overlap for 2 seconds, each frame is used as an independent sample mel-frequency spectrum coefficient characteristic for classification by a classifier (a convolution neural network is specifically adopted in the embodiment), and 126875 samples are obtained as a marked underwater sound signal sample set L; the remaining 101500 samples constitute unlabeled sample set U. The method for classifying underwater sound signals according to the present invention for small sample problem will be described in further detail with respect to the noted underwater sound signal sample set L composed of 126875 samples and the unlabeled sample set U composed of the remaining 101500 samples.
As shown in fig. 1, the underwater sound signal classification method for the small sample problem of the present embodiment includes:
s1, randomly sampling a marked underwater sound signal sample set L;
s2, training classifier C based on newly constructed positive and negative sample sets of random sampling s ；
S3, using classifier C s Carrying out classification probability prediction on the unlabeled sample set U and carrying out sample migration;
s4, training an initial classifier C based on newly constructed positive and negative sample sets of sample migration 0 And calculates an initial classifier C 0 Corresponding toIndex value F 0 The method comprises the steps of carrying out a first treatment on the surface of the Initializing the value of an iteration counter i;
s5, using the i-1 th classifier C i-1 Carrying out classification probability prediction on the current positive and negative sample sets and carrying out sample migration;
s6, training an ith classifier C based on newly constructed positive and negative sample sets of sample migration i And calculates the ith classifier C i Corresponding index value F i ；
S7, if the ith classifier C i Index value F of (2) i Greater than the i-1 th classifier C i-1 Index value F of (2) i-1 The iteration counter i is increased by 1, and the step S5 is continued to iterate; otherwise, the i-1 th classifier C i-1 As an optimal classifier for classifying unlabeled underwater acoustic signals.
In this embodiment, step S1 includes: dividing the marked underwater sound signal sample set L into an abnormal underwater sound signal sample set L P And a set L of normal underwater sound signal samples N In a given quantitative proportion r p From set L P Randomly selected part of the samples form a set S P In a given quantitative proportion r n From set L N Randomly selected part of the samples form a set S N . In this embodiment, r p =0.02 and r n =0.05, the marked underwater sound signal sample set L contains 25375 samples, of which 6732 abnormal underwater sound signal samples and 18643 normal underwater sound signal samples are in proportion to the number r p =0.05 slave set L P Randomly selecting part of abnormal underwater sound signal samples to form a set S P ，S P The number of samples isAccording to the quantity proportion r n =0.05 slave set L N Randomly selecting part of normal underwater sound signal samples to form a set S N ，S N The number of samples is
In the present embodiment of the present invention,the step S2 comprises the following steps: using L P -S P As a positive sample set and given tag 1, u+s is used P +S N As a negative sample set and given a label 0, training the positive sample set and the negative sample set by using a supervised learning method to obtain a classifier C s Wherein U is an unlabeled sample set, and the unlabeled sample set contains unlabeled underwater sound signal samples. It should be noted that, classifier C s Initial classifier C 0 Ith classifier C i The required well-known classification model can be adopted according to the needs, and the input of the classification model can be selected to directly take the sample as the input of the classification model according to the needs, or the sample characteristics are extracted from the sample as the input of the classification model. As an alternative implementation manner, the present embodiment uses mel-frequency cepstrum coefficient features commonly used for voice recognition as input of a classification model for the underwater sound signal.
In this embodiment, step S3 includes: for set S P All samples in (1) use classifier C s Predicting the probability of the abnormal underwater sound signal sample, and recording the maximum abnormal probability value t p,max And a minimum anomaly probability value t p,min The method comprises the steps of carrying out a first treatment on the surface of the For all samples in unlabeled sample set U, classifier C is used s Predicting the probability of being an abnormal underwater sound signal sample and combining the maximum abnormal probability value t p,max And a minimum anomaly probability value t p,min Migrate samples in unlabeled sample set U (unlabeled sample set U in this embodiment contains 20000 samples) to set L P And L N In (a) and (b); for set S N All samples in (1) use classifier C s Predicting the probability of the signal sample being a normal underwater sound signal sample, and recording the maximum normal probability value t n,max And a minimum normal probability value t n,min The method comprises the steps of carrying out a first treatment on the surface of the For all samples in unlabeled sample set U, classifier C is used s Predicting the probability of the signal sample being a normal underwater sound signal sample, and combining the maximum normal probability value t n,max And a minimum normal probability value t n,min Migrating samples in unlabeled sample set U to set L P And L N Is a kind of medium. Wherein pair sets S P All of the samples in (a)The using classifier C s Predicting the probability of the abnormal underwater sound signal sample, and maximizing the abnormal probability value t p,max And a minimum anomaly probability value t p,min Can be expressed as:
t p,max ＝max(P r (y＝1|x∈S p ))，
t p,min ＝min(P r (y＝1|x∈S p ))，
in the above formula, max represents maximum value, min represents minimum value, and y represents classifier C s X represents the classifier C s Input (set S) P Samples of (1)), P) r (y＝1|x∈S p ) A probability value representing an output of 1.
In the present embodiment, the maximum abnormal probability value t is combined p,max And a minimum anomaly probability value t p,min Migrating samples in unlabeled sample set U to abnormal underwater sound signal sample set L P And a normal underwater sound signal sample set L N Comprises the following steps: for all s in unlabeled sample set U p1 The probability value is greater than or equal to the maximum abnormal probability value t p,max Will be s above p1 The individual samples are removed from the unlabeled sample set U and added to the abnormal underwater sound signal sample set L P The method comprises the steps of carrying out a first treatment on the surface of the For all s in unlabeled sample set U p2 The probability value is smaller than the maximum abnormal probability value t p,max And is greater than the minimum abnormal probability value t p,min If s is satisfied p2 ≤c p *s p1 Then s is as described above p2 The individual samples are removed from the unlabeled sample set U and added to the normal underwater sound signal sample set L N If s p2 >c p *s p1 Then will first go above s p2 Sorting the samples from big to small according to probability value, and then sorting the previous c p *s p1 The individual samples are removed from the unlabeled sample set U and added to the abnormal underwater sound signal sample set L P The method comprises the steps of carrying out a first treatment on the surface of the Wherein c p Is a constant parameter. In this embodiment there is c p =0.12, for all probability values less than t p,max And is greater than t p,min Record the number of samples s p2 If s p2 ≤0.12*s p1 Then all s will be p2 The individual samples are removed from set U and added to set L N If s p2 >0.12*s p1 Then for all s p2 The samples are sorted from big to small according to probability values and then the previous c p *s p1 The individual samples are removed from set U and added to set L P 。
In this embodiment, the maximum normal probability value t is combined n,max And a minimum normal probability value t n,min Migrating samples in unlabeled sample set U to abnormal underwater sound signal sample set L P And a normal underwater sound signal sample set L N Comprises the following steps: for all s in unlabeled sample set U n1 The probability value is greater than or equal to the maximum normal probability value t n,max Will be s above n1 The individual samples are removed from the unlabeled sample set U and added to the normal underwater sound signal sample set L N The method comprises the steps of carrying out a first treatment on the surface of the For all s in unlabeled sample set U n2 The probability values are smaller than the maximum normal probability value t n,max And is greater than the minimum normal probability value t n,min If s is satisfied n2 ≤c n *s n1 Then s is as described above n2 The individual samples are removed from the unlabeled sample set U and added to the normal underwater sound signal sample set L N If s n2 >c n *s n1 Then will first go above s n2 Sorting the samples from big to small according to probability value, and then sorting the previous c n *s n1 The individual samples are removed from the unlabeled sample set U and added to the normal underwater sound signal sample set L N The method comprises the steps of carrying out a first treatment on the surface of the Wherein c n Is a constant parameter. Maximum normal probability value t n,max And a minimum normal probability value t n,min Can be expressed as:
t n,max ＝max(P r (y＝0|x∈S n ))，
t n,min ＝min(P r (y＝0|x∈S n ))，
in the above formula, max represents maximum value, min represents minimum value, and y represents classifier C s X represents the classifier C s Input (set S) n Samples of (1)), P) r (y＝0|x∈S n ) Representing a probability value of 0 output. C in this embodiment n =0.1, then there is a pair ofAt all probability values less than t n,max And is greater than t n,min Record the number of samples s n2 If s n2 ≤0.1*s n1 Then all s will be n2 The individual samples are removed from set U and added to set L N If s n2 >0.1*s n1 Then for all s n2 The samples are sorted from big to small according to probability values and then the previous c n *s n1 The individual samples are removed from set U and added to set L N 。
In this embodiment, step S4 trains the initial classifier C based on the positive and negative sample sets newly constructed by sample migration 0 Comprising the following steps: using set L P As a positive sample set and given tag 1, set L is used N As a negative sample set and given a label 0, training by using a supervised learning method by adopting the positive sample set and the negative sample set to obtain an initial classifier C 0 。
In this embodiment, step S4 further includes establishing a set L PA And L NA The two incremental training sets are initially empty, and step S5 includes: for set L P Using the i-1 th classifier C i-1 Predicting the probability of the abnormal underwater sound signal sample, and recording the minimum probability value u p This can be expressed as:
u p ＝min(P r (y＝1|x∈L P ))，
in the above formula, max represents maximum value, min represents minimum value, and y represents i-1 th classifier C i-1 X represents the i-1 th classifier C i-1 Input (set L) P Samples of (1)), P) r (y＝1|x∈L P ) A probability value representing an output of 1.
For all samples in unlabeled sample set U, use the i-1 st classifier C i-1 Predicting the probability of the abnormal underwater sound signal sample, and making the probability value larger than the minimum probability value u p Samples removed from unlabeled sample set U and added to set L PA The method comprises the steps of carrying out a first treatment on the surface of the For set L N Using the i-1 th classifier C i-1 Predicting the probability of the signal sample being a normal underwater sound signal sample, and recording the minimumProbability value u n This can be expressed as:
u n ＝min(P r (y＝0|x∈L N ))，
in the above formula, max represents maximum value, min represents minimum value, and y represents i-1 th classifier C i-1 X represents the i-1 th classifier C i-1 Input (set L) N Samples of (1)), P) r (y＝0|x∈L N ) Representing a probability value of 0 output.
For all samples in unlabeled sample set U, use the i-1 st classifier C i-1 Predicting the probability of the normal underwater sound signal sample, and making the probability value larger than the minimum probability value u n Samples are removed from set U and added to set L NA The method comprises the steps of carrying out a first treatment on the surface of the The step S6 comprises the following steps: using L P +L PA As a positive sample set and given tag 1, use L N +L NA As a negative sample set and given a label 0, training by using a supervised learning method by adopting the positive sample set and the negative sample set to obtain an ith classifier C i The method comprises the steps of carrying out a first treatment on the surface of the At set L P And L N The ith classifier C obtained by the previous calculation of the iteration of the round i Index value F of (2) i The method comprises the steps of carrying out a first treatment on the surface of the When the iteration counter i is incremented by 1 in step S7, the method further comprises the step of collecting L PA Join set L P Will be set L NA Join set L N 。
Finally, the i-1 th classifier C i-1 As an optimal classifier, can be used to classify untagged underwater acoustic signals. For example, as an alternative implementation, the classifier C in this embodiment i-1 As an optimal classifier, L PA And L NA Adding U, and using classifier C i-1 The samples in unlabeled sample set U are classified. Specifically, the method of the embodiment iterates the optimal classifier C obtained after 13 rounds 12 The classification accuracy for unlabeled exemplar set U was 76.2%. The control group adopts a convolution network as a classifier, the labeled sample set L is used for training, and the classification accuracy test adopts layered 2×5 fold cross validation. Equally dividing training samples into 5 parts, wherein the proportion of 4 types of samples in each part is equal, each part is taken as a test set in turn, and the rest isTraining and verifying the set, wherein the sample ratio of each test training, verifying and testing set is 7:1:2, and the total number of the tests is 10, the training set is used for model training, the verifying set is used for judging the time for stopping training, and the classifier obtained by training is evaluated on the testing set. The classifier with the highest test classification accuracy is selected as the optimal classifier by the control group, and the classification accuracy of the unlabeled sample set U is 71.9%. The test results of the embodiment show that compared with the traditional method, the underwater sound signal classification method for the small sample problem can improve the classification precision of the abnormal underwater sound signal sample.
In summary, the underwater sound signal classification method for small sample problem of the present embodiment includes randomly sampling the marked underwater sound signal sample set and training the classifier C s The method comprises the steps of carrying out a first treatment on the surface of the Using classifier C s Carrying out classification probability prediction and sample migration on unlabeled sample sets, training an initial classifier of the constructed positive and negative sample sets, and calculating index values; and carrying out classification probability prediction on the current positive and negative sample set by using the i-1 th classifier, carrying out sample migration, training the i-th classifier based on the newly constructed positive and negative sample set, calculating an index value, and selecting whether to continue iteration according to the index value. The method of the embodiment converts the small sample problem into the traditional supervised and semi-supervised learning problem, digs reliable normal underwater sound signal samples and abnormal underwater sound signal samples from unlabeled sample sets by means of marked samples, expands the scale of the marked sample set for training a machine learning model, and accordingly improves the classification accuracy of the abnormal underwater sound signal samples.
In addition, the embodiment also provides a small sample problem-oriented underwater sound signal classification system, which comprises a microprocessor and a memory which are connected with each other, wherein the microprocessor is programmed or configured to execute the small sample problem-oriented underwater sound signal classification method. The present embodiment also provides a computer readable storage medium having stored therein a computer program for being programmed or configured by a microprocessor to perform the small sample problem oriented underwater sound signal classification method.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein. The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above description is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above examples, and all technical solutions belonging to the concept of the present invention belong to the protection scope of the present invention. It should be noted that modifications and adaptations to the present invention may occur to one skilled in the art without departing from the principles of the present invention and are intended to be within the scope of the present invention.

Claims (10)

1. A method for classifying underwater acoustic signals for small sample problems, comprising:
s1, randomly sampling a marked underwater sound signal sample set L;
s2, training classifier C based on newly constructed positive and negative sample sets of random sampling s ；
S3, using classifier C s Carrying out classification probability prediction on the unlabeled sample set U and carrying out sample migration;
s4, training an initial classifier C based on newly constructed positive and negative sample sets of sample migration 0 And calculates an initial classifier C 0 Corresponding index value F 0 The method comprises the steps of carrying out a first treatment on the surface of the Initializing the value of an iteration counter i;
s5, using the i-1 th classifier C i-1 Carrying out classification probability prediction on the current positive and negative sample sets and carrying out sample migration;
s6, training an ith classifier C based on newly constructed positive and negative sample sets of sample migration i And calculates the ith classifier C i Corresponding index value F i ；
S7, if the ith classifier C i Index value F of (2) i Greater than the i-1 th classifier C i-1 Index value F of (2) i-1 The iteration counter i is increased by 1, and the step S5 is continued to iterate; otherwise, the i-1 th classifier C i-1 As an optimal classifier for classifying unlabeled underwater acoustic signals.
2. The method for classifying underwater acoustic signals for small sample problems as claimed in claim 1, wherein the step S1 comprises: dividing the marked underwater sound signal sample set L into an abnormal underwater sound signal sample set L P And a set L of normal underwater sound signal samples N In a given quantitative proportion r p From the collectionL is combined with P Randomly selected part of the samples form a set S P In a given quantitative proportion r n From set L N Randomly selected part of the samples form a set S N 。
3. The method for classifying underwater acoustic signals for small sample problems as claimed in claim 2, wherein the step S2 comprises: using L P -S P As a positive sample set and given tag 1, u+s is used P +S N As a negative sample set and given a label 0, training the positive sample set and the negative sample set by using a supervised learning method to obtain a classifier C s Wherein U is an unlabeled sample set, and the unlabeled sample set contains unlabeled underwater sound signal samples.
4. The method for classifying underwater acoustic signals for small sample problems as claimed in claim 2, wherein the step S3 comprises: for set S P All samples in (1) use classifier C s Predicting the probability of the abnormal underwater sound signal sample, and recording the maximum abnormal probability value t p，max And a minimum anomaly probability value t p，min The method comprises the steps of carrying out a first treatment on the surface of the For all samples in unlabeled sample set U, classifier C is used s Predicting the probability of being an abnormal underwater sound signal sample and combining the maximum abnormal probability value t p，max And a minimum anomaly probability value t p，min Migrating samples in unlabeled sample set U to set L P And L N In (a) and (b); for set S N All samples in (1) use classifier C s Predicting the probability of the signal sample being a normal underwater sound signal sample, and recording the maximum normal probability value t n，max And a minimum normal probability value t n，min The method comprises the steps of carrying out a first treatment on the surface of the For all samples in unlabeled sample set U, classifier C is used s Predicting the probability of the signal sample being a normal underwater sound signal sample, and combining the maximum normal probability value t n，max And a minimum normal probability value t n，min Migrating samples in unlabeled sample set U to set L P And L N Is a kind of medium.
5. The small sample problem oriented underwater sound signal classification method according to claim 4, wherein said combining maximum abnormal probability value t p，max And a minimum anomaly probability value t p，min Migrating samples in unlabeled sample set U to abnormal underwater sound signal sample set L P And a normal underwater sound signal sample set L N Comprises the following steps: for all s in unlabeled sample set U p1 The probability value is greater than or equal to the maximum abnormal probability value t p，max Will be s above p1 The individual samples are removed from the unlabeled sample set U and added to the abnormal underwater sound signal sample set L P The method comprises the steps of carrying out a first treatment on the surface of the For all s in unlabeled sample set U p2 The probability value is smaller than the maximum abnormal probability value t p，max And is greater than the minimum abnormal probability value t p，min If s is satisfied p2 ≤c p *s p1 Then s is as described above p2 The individual samples are removed from the unlabeled sample set U and added to the normal underwater sound signal sample set L N If s p2 ＞c p *s p1 Then will first go above s p2 Sorting the samples from big to small according to probability value, and then sorting the previous c p *s p1 The individual samples are removed from the unlabeled sample set U and added to the abnormal underwater sound signal sample set L P The method comprises the steps of carrying out a first treatment on the surface of the Wherein c p Is a constant parameter.
6. The method for classifying underwater acoustic signals for small sample problems as claimed in claim 4, wherein said combined maximum normal probability value t n，max And a minimum normal probability value t n，min Migrating samples in unlabeled sample set U to abnormal underwater sound signal sample set L P And a normal underwater sound signal sample set L N Comprises the following steps: for all s in unlabeled sample set U n1 The probability value is greater than or equal to the maximum normal probability value t n，max Will be s above n1 The individual samples are removed from the unlabeled sample set U and added to the normal underwater sound signal sample set L N The method comprises the steps of carrying out a first treatment on the surface of the For all s in unlabeled sample set U n2 The probability values are smaller than the maximum normal probability value t n，max And is greater than the minimum normal probability value t n，min If s is satisfied n2 ≤c n *s n1 Then s is as described above n2 The individual samples are removed from the unlabeled sample set U and added to the normal underwater sound signal sample set L N If s n2 ＞c n *s n1 Then will first go above s n2 Sorting the samples from big to small according to probability value, and then sorting the previous c n *s n1 The individual samples are removed from the unlabeled sample set U and added to the normal underwater sound signal sample set L N The method comprises the steps of carrying out a first treatment on the surface of the Wherein c n Is a constant parameter.
7. The small sample problem oriented underwater sound signal classification method according to claim 2, wherein the initial classifier C is trained based on the positive and negative sample sets newly constructed by sample migration in step S4 0 Comprising the following steps: using set L P As a positive sample set and given tag 1, set L is used N As a negative sample set and given a label 0, training by using a supervised learning method by adopting the positive sample set and the negative sample set to obtain an initial classifier C 0 。
8. The method of classifying underwater acoustic signals for small sample problems as claimed in claim 2, wherein the step S4 further comprises creating a set L PA And L NA The two incremental training sets are initially empty, and step S5 includes: for set L P Using the i-1 th classifier C i-1 Predicting the probability of the abnormal underwater sound signal sample, and recording the minimum probability value u p The method comprises the steps of carrying out a first treatment on the surface of the For all samples in unlabeled sample set U, use the i-1 st classifier C i-1 Predicting the probability of the abnormal underwater sound signal sample, and making the probability value larger than the minimum probability value u p Samples removed from unlabeled sample set U and added to set L PA The method comprises the steps of carrying out a first treatment on the surface of the For set L N Using the i-1 th classifier C i-1 Predicting the probability of the signal sample being a normal underwater sound signal sample, and recording the minimum probability value u n The method comprises the steps of carrying out a first treatment on the surface of the For all samples in unlabeled sample set U, use the i-1 st classifier C i-1 For which it is a normal water sound messagePredicting the probability of the number sample, wherein the probability value is larger than the minimum probability value u n Samples are removed from set U and added to set L NA The method comprises the steps of carrying out a first treatment on the surface of the The step S6 comprises the following steps: using L P +L PA As a positive sample set and given tag 1, use L N +L NA As a negative sample set and given a label 0, training by using a supervised learning method by adopting the positive sample set and the negative sample set to obtain an ith classifier C i The method comprises the steps of carrying out a first treatment on the surface of the At set L P And L N The ith classifier C obtained by the previous calculation of the iteration of the round i Index value F of (2) i The method comprises the steps of carrying out a first treatment on the surface of the When the iteration counter i is incremented by 1 in step S7, the method further comprises the step of collecting L PA Join set L P Will be set L NA Join set L N 。
9. A small sample problem oriented underwater acoustic signal classification system comprising a microprocessor and a memory interconnected, wherein the microprocessor is programmed or configured to perform the small sample problem oriented underwater acoustic signal classification method of any of claims 1-8.
10. A computer readable storage medium having a computer program stored therein, wherein the computer program is for being programmed or configured by a microprocessor to perform the small sample problem oriented underwater sound signal classification method of any of claims 1-8.
CN202311406647.XA 2023-10-26 2023-10-26 Underwater sound signal classification method, system and medium for small sample problem Pending CN117574262A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311406647.XA CN117574262A (en) 2023-10-26 2023-10-26 Underwater sound signal classification method, system and medium for small sample problem

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311406647.XA CN117574262A (en) 2023-10-26 2023-10-26 Underwater sound signal classification method, system and medium for small sample problem

Publications (1)

Publication Number Publication Date
CN117574262A true CN117574262A (en) 2024-02-20

Family

ID=89883365

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311406647.XA Pending CN117574262A (en) 2023-10-26 2023-10-26 Underwater sound signal classification method, system and medium for small sample problem

Country Status (1)

Country Link
CN (1) CN117574262A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118098221A (en) * 2024-04-23 2024-05-28 武汉理工大学三亚科教创新园 Small sample ocean sound event detection method based on self-adaptive learning

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118098221A (en) * 2024-04-23 2024-05-28 武汉理工大学三亚科教创新园 Small sample ocean sound event detection method based on self-adaptive learning

Similar Documents

Publication Publication Date Title
CN108984724B (en) Method for improving emotion classification accuracy of specific attributes by using high-dimensional representation
CN111914644B (en) Dual-mode cooperation based weak supervision time sequence action positioning method and system
CN109583501B (en) Method, device, equipment and medium for generating image classification and classification recognition model
CN109816032B (en) Unbiased mapping zero sample classification method and device based on generative countermeasure network
CN112069310B (en) Text classification method and system based on active learning strategy
CN107945210B (en) Target tracking method based on deep learning and environment self-adaption
CN112685504B (en) Production process-oriented distributed migration chart learning method
CN111653275B (en) Method and device for constructing voice recognition model based on LSTM-CTC tail convolution and voice recognition method
CN110851176B (en) Clone code detection method capable of automatically constructing and utilizing pseudo-clone corpus
CN110879881B (en) Mouse track recognition method based on feature component hierarchy and semi-supervised random forest
CN117574262A (en) Underwater sound signal classification method, system and medium for small sample problem
CN110704616B (en) Equipment alarm work order identification method and device
CN111126576A (en) Novel training strategy for deep learning
CN109740481B (en) CNN and LSTM combined atrial fibrillation signal classification system based on jump connection
CN111160959A (en) User click conversion estimation method and device
CN116975776A (en) Multi-mode data fusion method and device based on tensor and mutual information
CN115812210A (en) Method and apparatus for enhancing performance of machine learning classification tasks
CN114841151A (en) Medical text entity relation joint extraction method based on decomposition-recombination strategy
Fonseca et al. Model-agnostic approaches to handling noisy labels when training sound event classifiers
CN117521063A (en) Malicious software detection method and device based on residual neural network and combined with transfer learning
CN113343123A (en) Training method and detection method for generating confrontation multiple relation graph network
CN107229944B (en) Semi-supervised active identification method based on cognitive information particles
CN115936003A (en) Software function point duplicate checking method, device, equipment and medium based on neural network
CN114896402A (en) Text relation extraction method, device, equipment and computer storage medium
CN114547264A (en) News diagram data identification method based on Mahalanobis distance and comparison learning

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