Underwater target identification method and device and computer readable storage medium

Abstract

The invention belongs to the technical field of target recognition, and relates to an underwater target recognition method and device and a computer readable storage medium. Performing modal decomposition on the underwater acoustic signal to obtain a plurality of intrinsic modal signals; inputting the plurality of intrinsic mode signals into a target identification model, and outputting a plurality of underwater target prediction category probability distributions; merging the plurality of underwater target prediction category probability distributions to obtain a new underwater target prediction category probability distribution; obtaining weighted average probability distribution based on the probability distribution of all underwater target prediction categories; calculating an average similarity and a fusion evidence similarity based on the prediction category probability distribution and the weighted average probability distribution of each underwater target; if the fusion evidence similarity is smaller than or equal to the average similarity, the intrinsic mode signals of the underwater acoustic signals are screened, and underwater target recognition is carried out based on the screened target intrinsic mode signals; through screening the signal components of the underwater sound signals, the signal components containing error feature information are eliminated, and the underwater target identification precision is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target recognition, in particular to an underwater target recognition method, device and computer readable storage medium. BACKGROUND

[0002] Underwater target detection is a technology that analyzes collected underwater acoustic signals, obtains target signal characteristics through processing, and then judges and identifies underwater target categories based on human or machine networks, which plays a very important role. With the development of science and technology, the further exploration and exploitation of marine resources, and the changes in the marine environment, the underwater environment has become increasingly complex, posing a potential threat to the marine ecological environment.

[0003] The prior art provides an underwater target detection method based on underwater acoustic signal decomposition and feature analysis recognition, which acquires underwater acoustic signals, decomposes the underwater acoustic signals, constructs a simple or complex convolutional neural network, extracts and fuses features of each signal component obtained by decomposition, and finally outputs the underwater target category corresponding to the underwater acoustic signals. However, due to the extreme complexity of underwater acoustic signals, for example, when the underwater target is moving, its signal will be superimposed with instantaneous interference (sound reflection of suddenly passing seabed organisms or seabed rocks), which results in that part of the signal components obtained by decomposition do not simply carry effective features about the target category, but are not simply environmental noise, but false superposition of interference signals and useful signals, i.e. pseudo-feature information carriers, which carry feature information unrelated or even conflicting with the true features of the underwater target category. Unlike noise signals, this pseudo-feature carrier does not simply mask effective feature information, but provides false feature information, which misleads the convolutional neural network and thus outputs an incorrect target recognition result, reducing the accuracy of underwater target recognition.

[0004] In summary, the existing underwater target recognition method based on signal decomposition ignores that part of the signal components obtained by decomposing the underwater acoustic signals act as pseudo-feature carriers, which provide feature information conflicting with that provided by other signal components, thus leading to an incorrect recognition result based on false feature information when recognizing the underwater target, affecting the accuracy of the underwater target recognition result. SUMMARY

[0005] To this end, the technical problem to be solved by the present application is to overcome the problem in the prior art that the underwater target recognition method based on signal decomposition ignores that part of the signal components obtained by decomposing the underwater acoustic signals act as pseudo-feature carriers, which provide feature information conflicting with that provided by other signal components, thus leading to an incorrect recognition result based on false feature information when recognizing the underwater target, affecting the accuracy of the underwater target recognition result.

[0006] To solve the above technical problems, the present application provides an underwater target recognition method, comprising: S10: modal decomposition is performed on the underwater acoustic signal to obtain a plurality of intrinsic mode signals; S20: the plurality of intrinsic mode signals are input into a target recognition model respectively, a plurality of underwater target predicted category probability distributions are output, a probability distribution set is obtained, the plurality of underwater target predicted category probability distributions are combined, a new underwater target predicted category probability distribution is obtained, and the new underwater target predicted category probability distribution is added to the probability distribution set; S30: the sum of the correlation coefficients of each underwater target predicted category probability distribution in the probability distribution set and the remaining underwater target predicted category probability distributions is taken as the support degree of each underwater target predicted category probability distribution; based on the underwater target predicted category probability distributions in the probability distribution set and the support degree weights thereof, a weighted average probability distribution is obtained; S40: the similarity between the underwater target predicted category probability distribution of each intrinsic mode signal and the weighted average probability distribution is taken as the evidence similarity of each intrinsic mode signal; based on the mean of the evidence similarities of all intrinsic mode signals, an average similarity is obtained; the similarity between the new underwater target predicted category probability distribution and the weighted average probability distribution is taken as the fusion evidence similarity; S50: the fusion evidence similarity and the average similarity are compared, if the fusion evidence similarity is less than or equal to the average similarity, the plurality of intrinsic mode signals of the underwater acoustic signal are screened, and step S20 is returned to be executed until the fusion evidence similarity is greater than the average similarity, and underwater target recognition is performed based on the target intrinsic mode signals obtained by screening.

[0007] Preferably, screening the plurality of intrinsic mode signals comprises: A plurality of underwater acoustic signal samples are obtained in the underwater acoustic signal data set of each category of underwater target to obtain a subset of underwater acoustic signal samples of the category; and a plurality of intrinsic mode signals of each underwater acoustic signal sample in all subsets of underwater acoustic signal samples are obtained. The Fisher score method is used to calculate the discrimination performance score of each intrinsic mode signal for different categories of underwater targets; The N intrinsic mode signals with the highest discrimination performance scores are taken as target intrinsic mode signals, so that the target intrinsic mode signals are screened from the plurality of intrinsic mode signals of the underwater acoustic signal.

[0008] Preferably, the Fisher score method is used to calculate the discrimination performance score of each intrinsic mode signal for different categories of underwater targets, comprising: The mean of the i-th intrinsic mode signal of all underwater acoustic signal samples in all subsets of underwater acoustic signal samples is calculated to obtain the global mean of the i-th intrinsic mode signal; Calculate the mean of the i-th intrinsic mode signal of each underwater acoustic signal sample in each underwater acoustic signal sample subset, and obtain the intra-class mean of the i-th intrinsic mode signal in each underwater acoustic signal sample subset; The inter-class dispersion of the i-th intrinsic mode signal is obtained by summing the squared differences between the intra-class mean and the global mean of the i-th intrinsic mode signal in each underwater acoustic signal sample subset. The intra-class dispersion of the i-th intrinsic mode signal is obtained by summing the squared differences between the i-th intrinsic mode signal and the intra-class mean of the i-th intrinsic mode signal in each underwater acoustic signal sample subset. Based on the ratio of inter-class dispersion to intra-class dispersion of the i-th intrinsic mode signal, the discrimination performance score of the i-th intrinsic mode signal for different categories of underwater targets is obtained.

[0009] Preferably, the support weight of each underwater target prediction category probability distribution is the ratio of the support of that underwater target prediction category probability distribution to the sum of the support of all underwater target prediction category probability distributions.

[0010] Preferably, the formula for calculating the correlation coefficient between the probability distributions of underwater target prediction categories is as follows: , in, Indicates the first The probability distribution of the predicted category of the first underwater target is related to the probability distribution of the second underwater target prediction category. The correlation coefficient between the probability distributions of the predicted categories of underwater targets; Indicates the first In the probability distribution of the underwater target prediction category, the th... The probability values ​​of each predicted category; Indicates the number of categories of underwater targets; Indicates the first In the probability distribution of the underwater target prediction category, the th... The probability values ​​of each predicted category; The formula for calculating the support of the probability distribution of underwater target prediction categories is: , in, Indicates the first Support of the probability distribution of the predicted categories of underwater targets; Indicates the number of intrinsic mode signals; This represents the number of probability distributions for underwater target prediction categories in the probability distribution set; The formula for calculating the weighted average probability distribution is: , in, represents a weighted average probability distribution; represents a support weight of the th underwater target prediction class probability distribution; represents the th underwater target prediction class probability distribution.

[0011] Preferably, the calculation formula of the evidence similarity of the intrinsic modal signal is: , wherein, represents the evidence similarity of the i th intrinsic modal signal, , represents the number of intrinsic modal signals; represents the th underwater target prediction class probability distribution, ; represents a weighted average probability distribution; represents a transpose; The calculation formula of the average similarity is: , wherein, represents the average similarity.

[0012] Preferably, the calculation formula of the global mean of the i th intrinsic modal signal is: , wherein, represents the global mean of the i th intrinsic modal signal; represents the number of subsets of underwater acoustic signal samples; represents the number of underwater acoustic signal samples in each subset of underwater acoustic signal samples; represents the i th intrinsic modal signal of the n th underwater acoustic signal sample in the c th subset of underwater acoustic signal samples; The calculation formula of the intra-class mean of the i th intrinsic modal signal in each subset of underwater acoustic signal samples is: , wherein, represents the intra-class mean of the i th intrinsic modal signal in the c th subset of underwater acoustic signal samples; The calculation formula of the inter-class dispersion of the i th intrinsic modal signal is: , wherein, represents the inter-class dispersion of the i th intrinsic modal signal; represents the weight of the squared difference between the intra-class mean and the global mean of the i th intrinsic modal signal in the c th subset of underwater acoustic signal samples. The calculation formula of the intra-class dispersion of the i-th intrinsic modal signal is: , Wherein, represents the intra-class dispersion of the i-th intrinsic modal signal; The calculation formula of the distinguishing performance score of the i-th intrinsic modal signal for different categories of underwater targets is: , Wherein, represents the distinguishing performance score of the i-th intrinsic modal signal for different categories of underwater targets.

[0013] Preferably, the underwater target recognition based on the target intrinsic modal signal screened out comprises: Inputting the target intrinsic modal signal screened out into a target recognition model respectively, and outputting the underwater target category prediction probability distribution of each target intrinsic modal signal; Merging the underwater target category prediction probability distribution of each target intrinsic modal signal by using the Dempster-Shafer fusion rule to obtain the underwater target category prediction comprehensive probability distribution, so as to obtain the underwater target category corresponding to the underwater acoustic signal.

[0014] The application further provides an underwater target recognition device, comprising: A signal decomposition module is configured to decompose the underwater acoustic signal into a plurality of intrinsic modal signals; A probability distribution acquisition module is configured to input the plurality of intrinsic modal signals into a target recognition model respectively, and output a plurality of underwater target prediction category probability distributions to obtain a probability distribution set; merge the plurality of underwater target prediction category probability distributions to obtain a new underwater target prediction category probability distribution, and add the new underwater target prediction category probability distribution to the probability distribution set; A probability distribution weighted fusion module is configured to take the sum of the correlation coefficients of each underwater target prediction category probability distribution and the remaining underwater target prediction category probability distributions in the probability distribution set as the support degree of each underwater target prediction category probability distribution; and obtain a weighted average probability distribution based on each underwater target prediction category probability distribution and the support degree weight thereof in the probability distribution set; A similarity calculation module is configured to take the similarity between the underwater target prediction category probability distribution of each intrinsic modal signal and the weighted average probability distribution as the evidence similarity of each intrinsic modal signal; obtain an average similarity based on the average of the evidence similarities of all intrinsic modal signals; and take the similarity between the new underwater target prediction category probability distribution and the weighted average probability distribution as the fusion evidence similarity; The signal screening and target recognition module is configured to compare the fusion evidence similarity and the average similarity, screen the plurality of intrinsic modal signals of the underwater acoustic signal if the fusion evidence similarity is less than or equal to the average similarity, and return to the step of executing the probability distribution acquisition module until the fusion evidence similarity is greater than the average similarity, and perform underwater target recognition based on the target intrinsic modal signal obtained through screening.

[0015] The application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the underwater target recognition method.

[0016] The underwater target recognition method provided by the application has the following beneficial effects: Since the feature information of the pseudo feature signal conflicts with the feature information of other signals, the prediction result based on the feature information is different from the prediction result based on other feature information, so the application first outputs a plurality of underwater target prediction category probability distributions based on the plurality of intrinsic modal signals obtained through decomposition of the underwater acoustic signal, takes each underwater target prediction category probability distribution as an original evidence source, and then screens the conflicting evidence sources and removes the corresponding intrinsic modal signals, so as to remove the pseudo feature signals in the plurality of intrinsic modal signals. Specifically, the comprehensive evidence source (i.e., a new underwater target prediction category probability distribution) that fuses all original evidence source situations is obtained by merging each underwater target prediction category probability distribution, and the weighted average evidence is calculated by using the original evidence source and the comprehensive evidence source. Since the comprehensive evidence source can reflect the situation after the fusion of all original evidences from the overall perspective, the weighted average evidence can better reflect the potential synergy or conflict relationship between the original evidence sources. At the same time, the fusion evidence similarity and the average similarity are compared to determine whether there is a conflicting evidence in the current original evidence source. Since the fusion evidence similarity represents the overall consistency of all original evidence sources with the weighted average evidence, the credibility of the comprehensive evidence can be more clearly reflected. When the fusion evidence similarity is low, it indicates that the conflicting evidence source has interfered with the overall evidence credibility, and the intrinsic modal signals are screened until the fusion evidence similarity is greater than the average similarity. At this time, the intrinsic modal signals obtained through screening are signal components containing effective feature information, so that more accurate underwater target recognition results can be output based on these signal components. The application introduces the evidence conflict theory to screen the signal components of the underwater acoustic signal, removes the signal components containing false feature information, so that the target recognition model is not disturbed by the false feature information, thereby improving the accuracy of underwater target recognition. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to make the content of the present application more easily understood, the present application will be further described in detail below according to specific embodiments of the present application and in conjunction with the accompanying drawings, in which: Figure 1 A flow chart of an underwater target recognition method provided by the present application is shown in Figure 2 A structural schematic diagram of an underwater target recognition device provided by the present application is shown in DETAILED DESCRIPTION

[0018] The present application will be further described below in conjunction with the accompanying drawings and specific embodiments, so that those skilled in the art can better understand the present application and implement it. However, the embodiments are not intended to limit the present application.

[0019] Please refer to Figure 1 , Figure 1 A flow chart of an underwater target recognition method provided by the present application is shown in the accompanying drawings, and the method specifically includes: S10: Modal decomposition is performed on the underwater acoustic signal to obtain a plurality of intrinsic mode signals.

[0020] S20: The plurality of intrinsic mode signals are respectively input into a target recognition model to output a plurality of underwater target predicted class probability distributions, and a probability distribution set is obtained; the plurality of underwater target predicted class probability distributions are combined to obtain a new underwater target predicted class probability distribution, and the new underwater target predicted class probability distribution is added to the probability distribution set.

[0021] For example, the Dempster-Shafer fusion rule can be used to combine the plurality of underwater target predicted class probability distributions, and the formula is expressed as: , wherein, represents the new underwater target predicted class probability distribution; represents the underwater target predicted class probability distribution corresponding to the Ith intrinsic mode signal.

[0022] S30: The sum of the correlation coefficients of each underwater target predicted class probability distribution in the probability distribution set and the remaining underwater target predicted class probability distributions is taken as the support degree of each underwater target predicted class probability distribution; and based on the underwater target predicted class probability distributions in the probability distribution set and the support degree weights thereof, a weighted average probability distribution is obtained.

[0023] S40: similarity between the underwater target prediction class probability distribution of each intrinsic modal signal and the weighted average probability distribution is taken as the evidence similarity of each intrinsic modal signal; an average similarity is obtained based on the average of the evidence similarities of all intrinsic modal signals; similarity between the new underwater target prediction class probability distribution and the weighted average probability distribution is taken as the fusion evidence similarity.

[0024] S50: the fusion evidence similarity is compared with the average similarity, if the fusion evidence similarity is less than or equal to the average similarity, the multiple intrinsic modal signals of the underwater acoustic signal are screened, and step S20 is returned to be executed until the fusion evidence similarity is greater than the average similarity, and the target intrinsic modal signal obtained through screening is used for underwater target recognition.

[0025] Specifically, the underwater target prediction class probability distribution of the multiple intrinsic modal signals of the underwater acoustic signal is taken as the evidence source, when one evidence source is different from other evidence sources, the evidence source is considered as the conflict evidence, and the intrinsic modal signal corresponding to the conflict evidence source is removed, which is equivalent to removing the pseudo feature signal, so that the consistency of the remaining signals for the prediction result can be ensured, thereby improving the accuracy of underwater target recognition.

[0026] Further, when the conflict evidence is identified, all the underwater target prediction class probability distributions are taken as the original evidence sources, the original evidence sources are combined to obtain the comprehensive evidence source (i.e. the new underwater target prediction class probability distribution), and the weighted average evidence is calculated by using the original evidence sources and the comprehensive evidence source, since the comprehensive evidence source can reflect the situation after the fusion of all original evidence sources from the overall perspective, so that the weighted average evidence can better reflect the potential synergy or conflict relationship between the original evidence sources. Meanwhile, the fusion evidence similarity is compared with the average similarity to determine whether there is conflict evidence in the current original evidence source, since the fusion evidence similarity represents the overall consistency between all original evidence sources and the weighted average evidence, so that the reliability of the comprehensive evidence can be more clearly reflected, when the fusion evidence similarity is low, it indicates that the conflict evidence source interferes with the overall evidence reliability, and this conflict evidence identification method takes the comprehensive evidence source as a relatively stable conflict identification benchmark, thereby improving the reliability of the conflict evidence identification.

[0027] Specifically, the calculation formula of the correlation coefficient between the underwater target prediction class probability distributions in step S30 is: , wherein, represents the correlation coefficient between the i-th underwater target prediction class probability distribution and the j-th underwater target prediction class probability distribution. ​a correlation coefficient between the probability distribution of the predicted category of the underwater target and the probability distribution of the predicted category of the underwater target; represents the probability value of the th predicted category in the th predicted category probability distribution of the underwater target; represents the number of categories of the underwater target; represents the probability value of the th predicted category in the th predicted category probability distribution of the underwater target.

[0028] The calculation formula of the support of the predicted category probability distribution of the underwater target is: , wherein, represents the support of the th predicted category probability distribution of the underwater target; represents the number of intrinsic modal signals; represents the number of predicted category probability distributions of the underwater target in the probability distribution set; Further, the support weight of each predicted category probability distribution of the underwater target is the ratio of the support of the predicted category probability distribution of the underwater target to the sum of the supports of all predicted category probability distributions of the underwater target.

[0029] Specifically, the calculation formula of the support of the predicted category probability distribution of the underwater target is: , , wherein, represents the support of the th predicted category probability distribution of the underwater target; represents the support weight of the th predicted category probability distribution of the underwater target.

[0030] Further, the calculation formula of the weighted average probability distribution is: , wherein, represents the weighted average probability distribution; represents the support weight of the th predicted category probability distribution of the underwater target; represents the th predicted category probability distribution of the underwater target.

[0031] Further, the calculation formula of the evidence similarity of the intrinsic modal signal in step S40 is: , wherein, an evidence similarity representing an i-th intrinsic modal signal, , an intrinsic modal signal quantity; an i-th underwater target prediction class probability distribution, ; a weighted average probability distribution; a transpose.

[0032] The calculation formula of the average similarity is: , wherein, an average similarity.

[0033] Specifically, when the fused evidence similarity is less than or equal to the average similarity, it indicates that there is a conflict evidence source in the existing multiple evidence sources, and therefore the conflict evidence source needs to be removed. In some embodiments, a random screening method can be used in step S50 to screen the intrinsic modal signals without conflict evidence sources. However, the random screening method lacks theoretical guidance and is inefficient. Therefore, the present application proposes a method for screening intrinsic modal signals.

[0034] Specifically, the screening of the multiple intrinsic modal signals in step S50 includes: Step 1: obtaining multiple underwater acoustic signal samples in the underwater acoustic signal data set of each class of underwater target to obtain a subset of underwater acoustic signal samples of the class; and obtaining multiple intrinsic modal signals of each underwater acoustic signal sample in all subsets of underwater acoustic signal samples.

[0035] Step 2: calculating the discrimination performance score of each intrinsic modal signal for different classes of underwater targets by using Fisher score method.

[0036] Step 3: taking the N intrinsic modal signals with the highest discrimination performance scores as target intrinsic modal signals, so as to screen the target intrinsic modal signals from the multiple intrinsic modal signals of the underwater acoustic signal.

[0037] Further, step 2 includes: Step 2-1: calculating the mean value of the i-th intrinsic modal signal of all underwater acoustic signal samples in all subsets of underwater acoustic signal samples to obtain the global mean value of the i-th intrinsic modal signal.

[0038] Specifically, the calculation formula of the global mean value of the i-th intrinsic modal signal is: , wherein, a global mean value of the i-th intrinsic modal signal; a quantity of subsets of underwater acoustic signal samples; ​denotes the number of underwater acoustic signal samples in each subset of underwater acoustic signal samples; denotes the i-th intrinsic mode signal of the n-th underwater acoustic signal sample in the c-th subset of underwater acoustic signal samples.

[0039] Step 2-2: Calculate the mean of the i-th intrinsic mode signal of each underwater acoustic signal sample in each subset of underwater acoustic signal samples to obtain the intra-class mean of the i-th intrinsic mode signal in each subset of underwater acoustic signal samples.

[0040] Specifically, the calculation formula of the intra-class mean of the i-th intrinsic mode signal in each subset of underwater acoustic signal samples is: , wherein, denotes the intra-class mean of the i-th intrinsic mode signal in the c-th subset of underwater acoustic signal samples.

[0041] Step 2-3: Obtain the inter-class dispersion of the i-th intrinsic mode signal based on the sum of squared differences between the intra-class mean and the global mean of the i-th intrinsic mode signal in each subset of underwater acoustic signal samples.

[0042] Specifically, the calculation formula of the inter-class dispersion of the i-th intrinsic mode signal is: , wherein, denotes the inter-class dispersion of the i-th intrinsic mode signal; denotes the weight of the squared difference between the intra-class mean and the global mean of the i-th intrinsic mode signal in the c-th subset of underwater acoustic signal samples.

[0043] Step 2-4: Obtain the intra-class dispersion of the i-th intrinsic mode signal based on the sum of squared differences between the i-th intrinsic mode signal of each underwater acoustic signal sample in each subset of underwater acoustic signal samples and the intra-class mean of the i-th intrinsic mode signal in each subset of underwater acoustic signal samples.

[0044] Specifically, the calculation formula of the intra-class dispersion of the i-th intrinsic mode signal is: , wherein, denotes the intra-class dispersion of the i-th intrinsic mode signal.

[0045] Step 2-5: Obtain the performance score of the i-th intrinsic mode signal for distinguishing different classes of underwater targets based on the ratio of the inter-class dispersion and the intra-class dispersion of the i-th intrinsic mode signal.

[0046] Specifically, the calculation formula of the performance score of the i-th intrinsic mode signal for distinguishing different classes of underwater targets is: , wherein, represents the discrimination performance score of the i-th intrinsic modal signal for different categories of underwater targets.

[0047] Since the inter-class dispersion is used to measure the dispersion degree of different categories of underwater acoustic signal samples on the intrinsic modal signal, and the intra-class dispersion is used to measure the compactness of the same underwater acoustic signal sample on the intrinsic modal signal, therefore, when the inter-class dispersion of a certain intrinsic modal signal is greater, the intra-class dispersion is smaller, that is, the discrimination performance score is greater, it means that the intrinsic modal signal can reflect the difference of different categories of underwater acoustic signals, and the ability to distinguish different categories of underwater acoustic signals is stronger. Therefore, the present application selects the N intrinsic modal signals with the maximum discrimination performance score as the target intrinsic modal signals, and selects the target intrinsic modal signals of the underwater acoustic signal, so as to improve the discrimination degree of the target recognition model for the underwater acoustic signal.

[0048] Further, the underwater target recognition based on the selected target intrinsic modal signal comprises: inputting the selected target intrinsic modal signal into the target recognition model, and outputting the underwater target category prediction probability distribution of each target intrinsic modal signal.

[0049] merging the underwater target category prediction probability distribution of each target intrinsic modal signal by using the Dempster-Shafer fusion rule to obtain the underwater target category prediction comprehensive probability distribution, so as to obtain the underwater target category corresponding to the underwater acoustic signal.

[0050] The effectiveness of the above underwater target recognition method is illustrated by a specific example, and the marine underwater acoustic signal data set in the OUA data set is used in this example.

[0051] The marine underwater acoustic signal data set of the OUA data set is collected in a marine experiment, and the data set is obtained by a 128-element horizontal line array (HLA) dragged by a 400-ton receiving ship. The device records the noise signals emitted by two source ships with a weight of 100 tons and 500 tons respectively, and the source ships run at different distances of 2 to 20 kilometers from the receiving ship. The total duration of the data set is 14.5 hours, and contains four different categories of classification data: target A: signal generated by ship A; target B: signal generated by ship B; target C: signal generated by ship A and ship B; and non-target signal D.

[0052] The total signal duration is 38726 seconds, including 4567 seconds of background noise D, 4819 seconds of target A, 20220 seconds of target B and 9120 seconds of target C. In order to maintain data balance, we selected 4500 samples from each category for experiment. Given the relatively small size of the data set, the length of each data sample is set to 1 second.

[0053] The target recognition model in this example is a back propagation (BP) neural network consisting of an input layer of 32 neurons, a hidden layer of 64 neurons, and an output layer. The specific steps for underwater target recognition include: (1) Set up the data set: the test set is completely separated from the training set and the validation set, and the proportion is set to 7:1:2.

[0054] (2) Find the appropriate decomposition parameter K, and use the multivariate variational mode decomposition method to decompose the underwater acoustic signal: use real data to analyze the signal energy ratio under different decomposition levels, hide the first component, and draw a picture to observe the decomposition energy ratio of the remaining components.

[0055] Specifically, after analysis, the number of decomposition modes K is set to 5 for the best effect, and the gray correlation degree between each decomposition component and the original signal is calculated. The results show that the decomposed signal and the original signal show a high degree of similarity, and the decomposition process is effective.

[0056] (3) Preliminary prediction and merging: the multiple intrinsic mode signals obtained by decomposition are input into the target recognition model, and the underwater target prediction class probability distribution corresponding to each intrinsic mode signal is output; the multiple underwater target prediction class probability distributions are merged to obtain a new underwater target prediction class probability distribution.

[0057] (4) Conflict evidence acquisition and correction: based on all underwater target prediction class probability distributions, calculate the average similarity and fusion evidence similarity, compare the fusion evidence similarity and the average similarity, and determine whether there is conflict evidence in the multiple intrinsic mode signals. If there is, filter the multiple intrinsic mode signals until there is no conflict evidence in the filtered intrinsic mode signals.

[0058] (5) Target recognition: based on the filtered intrinsic mode signals without conflict evidence, perform underwater target recognition.

[0059] The accuracy of the validation set in this example is improved by 0.48% to 4.89%, and the accuracy of the test set is improved by 0.97% to 8.26%. As shown in Table 1, the data provided by this example before and after screening of different categories of underwater acoustic signal discrimination: Table 1

[0060] It is worth noting that after analyzing the underwater target recognition results before and after the intrinsic mode signal screening, this example found that the B, C, and D categories of underwater acoustic signals are more likely to be confused. After screening the intrinsic mode signals, the discrimination of these three categories of underwater acoustic signals is significantly improved, indicating that the method provided by the present application can improve the accuracy of underwater target recognition.

[0061] Based on the underwater target recognition method provided in the above embodiments, an embodiment of the present application further provides an underwater target recognition device, as shown in the figure, which specifically comprises: Figure 2 A signal decomposition module 10 is configured to perform modal decomposition on the underwater acoustic signal to obtain a plurality of intrinsic mode signals.

[0062] A probability distribution acquisition module 20 is configured to input the plurality of intrinsic mode signals into a target recognition model respectively, output a plurality of underwater target predicted category probability distributions, obtain a probability distribution set, and combine the plurality of underwater target predicted category probability distributions to obtain a new underwater target predicted category probability distribution and add the new underwater target predicted category probability distribution to the probability distribution set.

[0063] A probability distribution weighted fusion module 30 is configured to take the sum of the correlation coefficients of each underwater target predicted category probability distribution and the remaining underwater target predicted category probability distributions in the probability distribution set as the support degree of each underwater target predicted category probability distribution, and obtain a weighted average probability distribution based on each underwater target predicted category probability distribution in the probability distribution set and the support degree weight thereof.

[0064] A similarity calculation module 40 is configured to take the similarity between the underwater target predicted category probability distribution of each intrinsic mode signal and the weighted average probability distribution as the evidence similarity of each intrinsic mode signal, obtain an average similarity based on the average of the evidence similarities of all intrinsic mode signals, and take the similarity between the new underwater target predicted category probability distribution and the weighted average probability distribution as the fusion evidence similarity.

[0065] A signal screening and target recognition module 50 is configured to compare the fusion evidence similarity and the average similarity, if the fusion evidence similarity is less than or equal to the average similarity, screen the plurality of intrinsic mode signals of the underwater acoustic signal, and return to execute the step of the probability distribution acquisition module until the fusion evidence similarity is greater than the average similarity, and perform underwater target recognition based on the target intrinsic mode signal obtained through screening.

[0066] The embodiment of the present application further provides a computer readable storage medium, and the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the underwater target recognition method described above.

[0067] ​Those skilled in the art will appreciate that embodiments of the application can be devised for a method, a system, or a computer program product. Accordingly, the present application can be embodied in the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code.

[0068] The present application is described in reference to the 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 block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can 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 illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0069] These computer program instructions can 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 instructions which implement the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0070] These computer program instructions can 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 illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0071] Obviously, the above-described embodiments are only examples and are not intended to limit the present application. Based on the above description, those skilled in the art can make other variations and modifications of the present application without departing from the present application. Neither requiring nor intending to limit the present application to the exact forms of implementations shown and described, the application is to cover all modifications and variations as long as they come within the scope of the present application.

Claims

1. A method of underwater target recognition, characterized in that, Comprise: S10: modal decomposition is carried out to the underwater acoustic signal, and a plurality of intrinsic mode signals are obtained; S20: the plurality of intrinsic mode signals are respectively input into the target recognition model, and a plurality of underwater target predicted category probability distributions are output, a probability distribution set is obtained, the plurality of underwater target predicted category probability distributions are combined, a new underwater target predicted category probability distribution is obtained, and the new underwater target predicted category probability distribution is added to the probability distribution set; S30: the sum of the correlation coefficients of each underwater target predicted category probability distribution in the probability distribution set and the remaining underwater target predicted category probability distribution is used as the support degree of each underwater target predicted category probability distribution; Based on the underwater target predicted category probability distribution in the probability distribution set and its support degree weight, a weighted average probability distribution is obtained; S40: the similarity between the underwater target predicted category probability distribution of each intrinsic mode signal and the weighted average probability distribution is used as the evidence similarity of each intrinsic mode signal;Based on the average of the evidence similarity of all intrinsic mode signals, an average similarity is obtained;The similarity between the new underwater target predicted category probability distribution and the weighted average probability distribution is used as the fusion evidence similarity; S50: compare the fusion evidence similarity with the average similarity, if the fusion evidence similarity is less than or equal to the average similarity, the plurality of intrinsic mode signals of the underwater acoustic signal are screened, and step S20 is returned to be executed until the fusion evidence similarity is greater than the average similarity, the target intrinsic mode signal obtained by screening is used for underwater target recognition.

2. The underwater target recognition method according to claim 1, characterized in that, Screening the plurality of intrinsic mode signals includes: Obtain a plurality of underwater acoustic signal samples in each category of underwater target underwater acoustic signal data set, obtain the underwater acoustic signal sample subset of this category;Obtain the plurality of intrinsic mode signals of each underwater acoustic signal sample in all underwater acoustic signal sample subsets; The Fisher score method is used to calculate the discrimination performance score of each intrinsic mode signal for different categories of underwater targets; The N intrinsic mode signals with the highest discrimination performance score are used as the target intrinsic mode signal, so that the target intrinsic mode signal is screened from the plurality of intrinsic mode signals of the underwater acoustic signal.

3. The underwater target recognition method according to claim 2, characterized in that, The Fisher score method is used to calculate the discrimination performance score of each intrinsic mode signal for different categories of underwater targets, including: The mean value of the i th intrinsic mode signal of all underwater acoustic signal samples in all underwater acoustic signal sample subsets is calculated, and the global mean value of the i th intrinsic mode signal is obtained; The mean value of the i th intrinsic mode signal of each underwater acoustic signal sample in each underwater acoustic signal sample subset is calculated, and the within-class mean value of the i th intrinsic mode signal in each underwater acoustic signal sample subset is obtained; Based on the sum of the square differences between the within-class mean value and the global mean value of the i th intrinsic mode signal in each underwater acoustic signal sample subset, the between-class dispersion of the i th intrinsic mode signal is obtained; Based on the sum of the square differences between the i th intrinsic mode signal of each underwater acoustic signal sample in each underwater acoustic signal sample subset and the within-class mean value of the i th intrinsic mode signal in each underwater acoustic signal sample subset, the within-class dispersion of the i th intrinsic mode signal is obtained; The ratio of the inter-class dispersion degree of the i-th eigenmodal signal to the intra-class dispersion degree of the i-th eigenmodal signal is used to obtain a score of the i-th eigenmodal signal for distinguishing different classes of underwater targets.

4. The underwater target recognition method of claim 1, wherein, The support weight of each underwater target prediction class probability distribution is the ratio of the support of the underwater target prediction class probability distribution to the sum of the supports of all underwater target prediction class probability distributions.

5. The underwater target recognition method of claim 1, wherein, The correlation coefficient between underwater target prediction class probability distributions is calculated according to the following formula: ， wherein, denotes a correlation coefficient between the first underwater target prediction class probability distribution and the second underwater target prediction class probability distribution; denotes a probability value of the first prediction class in the first underwater target prediction class probability distribution; denotes a number of classes of underwater targets; denotes a probability value of the first prediction class in the first underwater target prediction class probability distribution; The support of the underwater target prediction class probability distribution is calculated according to the following formula: ， wherein, represents a support of the underwater target prediction class probability distribution; represents a number of intrinsic modal signals; represents a number of underwater target prediction class probability distributions in the set of probability distributions;​ The weighted average probability distribution is calculated according to the following formula: ， wherein, denotes a weighted average probability distribution; denotes a support weight of the th underwater target prediction class probability distribution; denotes the th underwater target prediction class probability distribution.

6. The underwater target recognition method of claim 1, wherein, The evidence similarity of the eigenmodal signal is calculated according to the following formula: ， in, This represents the evidence similarity of the i-th intrinsic mode signal. , Indicates the number of intrinsic mode signals; Indicates the first Probability distribution of underwater target prediction categories ; This represents a weighted average probability distribution; Indicates transpose; The average similarity is calculated according to the following formula: ， wherein, represents the average similarity.

7. The underwater target recognition method of claim 3, wherein, The global mean of the i-th eigenmodal signal is calculated according to the following formula: ， wherein, represents the global mean of the i-th eigenmodal signal; represents the number of subsets of hydroacoustic signal samples; represents the number of hydroacoustic signal samples in each subset of hydroacoustic signal samples; represents the i-th eigenmodal signal of the n-th hydroacoustic signal sample in the c-th subset of hydroacoustic signal samples; The intra-class mean of the i-th eigenmodal signal in each subset of underwater acoustic signal samples is calculated according to the following formula: ， wherein, represents the intra-class mean of the ith intrinsic modal signal in the cth subset of the underwater acoustic signal samples; The inter-class dispersion degree of the i-th eigenmodal signal is calculated according to the following formula: ， wherein, represents the between-class scatter of the i-th eigen-modal signal; represents the weight of the squared difference between the within-class mean and the global mean of the i-th eigen-modal signal in the c-th subset of underwater acoustic signal samples. The intra-class dispersion degree of the i-th eigenmodal signal is calculated according to the following formula: ， wherein, represents the intra-class dispersion of the i-th eigenmodal signal; The score of the i-th eigenmodal signal for distinguishing different classes of underwater targets is calculated according to the following formula: ， wherein, represents the performance score of the i-th intrinsic modal signal in distinguishing different categories of underwater targets.

8. The underwater target recognition method of claim 1, wherein, The underwater target recognition based on the screened target eigenmodal signal includes: The screened target eigenmodal signal is input into the target recognition model to output underwater target class prediction probability distributions of each target eigenmodal signal. The underwater target class prediction comprehensive probability distribution is obtained by merging the underwater target class prediction probability distributions of each target eigenmodal signal using the Dempster-Shafer fusion rule, thereby obtaining the underwater target class corresponding to the underwater acoustic signal.

9. An underwater target recognition device, characterized by It includes: The signal decomposition module is configured to perform modal decomposition on the underwater acoustic signal to obtain a plurality of eigenmodal signals. The probability distribution acquisition module is configured to input the plurality of eigenmodal signals into the target recognition model to output a plurality of underwater target prediction class probability distributions, obtain a probability distribution set, and merge the plurality of underwater target prediction class probability distributions to obtain a new underwater target prediction class probability distribution and add the new underwater target prediction class probability distribution to the probability distribution set. The probability distribution weighted fusion module is configured to use the sum of the correlation coefficients of each underwater target prediction class probability distribution and the remaining underwater target prediction class probability distributions in the probability distribution set as the support of each underwater target prediction class probability distribution. The weighted average probability distribution is obtained based on each underwater target prediction class probability distribution and its support weight in the probability distribution set. The similarity calculation module is configured to use the similarity between each eigenmodal signal and the weighted average probability distribution as the evidence similarity of each eigenmodal signal, obtain the average similarity based on the mean of the evidence similarity of all eigenmodal signals, and use the similarity between the new underwater target prediction class probability distribution and the weighted average probability distribution as the fusion evidence similarity. The signal screening and target recognition module is configured to compare the fusion evidence similarity with the average similarity, screen the multiple intrinsic mode signals of the underwater acoustic signal if the fusion evidence similarity is less than or equal to the average similarity, and return to the step of obtaining the probability distribution until the fusion evidence similarity is greater than the average similarity, and then recognize the underwater target based on the target intrinsic mode signal obtained through the screening.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the underwater target recognition method in any one of claims 1 to 8.

Images