A UUV self-noise separation method based on decomposition, classification and reconstruction
By using a decomposition, classification, and reconstruction method, and employing empirical mode decomposition and feature distance evaluation techniques, UUV self-noise is separated, thus solving the problem of UUV self-noise interference and improving signal reception and target detection performance.
Patent Information
- Application Number
- CN202410773884.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-17
- Publication Date
- 2025-11-21
- Estimated Expiration
- 2044-06-17
AI Technical Summary
UUV self-noise interference affects signal reception performance and target detection accuracy, and existing technologies struggle to effectively separate and mitigate self-noise from different sources.
The method of decomposition, classification and reconstruction is adopted. The empirical mode decomposition algorithm is used to decompose the self-noise signal source into multiple signal mode components, extract time domain and frequency domain features, screen sensitive feature sets through feature distance evaluation technology, and perform normalization and clustering processing to finally reconstruct the self-noise separation result.
Effectively separate UUV self-noise, suppress its impact on detection performance, and improve signal reception capability and target detection accuracy.
Smart Images
Figure CN118629420B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of underwater vehicle self-noise feature extraction, identification and separation technology, and in particular to a UUV self-noise separation method of decomposition, classification and reconstruction. Background Technology
[0002] Unmanned Underwater Vehicle (UUV) self-noise refers to various interference noises present at the signal receiver during underwater acoustic communication, target detection, and other processes, affecting the signal-to-noise ratio (SNR) of the received signal. The magnitude of UUV self-noise determines the performance indicators of the UUV acoustic payload and has become crucial for improving related technical performance. UUVs have complex structures, and the sources of their self-noise are also complex; different sources of self-noise have varying impacts on the acoustic payload system. Extracting and separating different types of self-noise from UUV self-noise and implementing targeted noise control are effective ways to improve UUV receiving capabilities.
[0003] However, in the actual research and application of UUV self-noise, the following problems exist:
[0004] Firstly, in the underwater environment, UUVs generally acquire external information through their onboard sensors and use underwater acoustic signals as the main medium for information transmission to achieve information interaction with the outside world. However, in practical applications, UUVs generally rely on the mobility and autonomy of their smaller unmanned platforms to carry out underwater missions such as submarine tracking and hunting, underwater search and investigation, maritime reconnaissance, and navigation assistance. However, due to the limitation of the unmanned platform's size, the underwater acoustic signal receiving system will be affected by the UUV's self-noise, thus affecting the signal reception performance.
[0005] Secondly, the self-noise of UUVs and the radiated noise of their targets have similar or even overlapping frequency bands. Passive sonar detects targets by receiving underwater signals, but the received signals are affected by the self-noise of UUVs, reducing the accuracy and sensitivity of target detection.
[0006] Therefore, it is necessary to propose a solution to improve one or more problems existing in the above-mentioned related technical solutions.
[0007] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention
[0008] This application provides a method for separating UUV self-noise through decomposition, classification, and reconstruction, including the following steps:
[0009] The self-noise signal source of the unmanned underwater vehicle is acquired, and the self-noise signal source is decomposed into multiple signal mode components using the empirical mode decomposition algorithm.
[0010] The time-domain and frequency-domain features of multiple signal mode components are extracted respectively to construct a time-domain feature set and a frequency-domain feature set, and the time-domain feature set and the frequency-domain feature set are combined into a comprehensive feature set;
[0011] The comprehensive feature set is filtered using the feature distance evaluation technique to obtain multiple sensitive feature sets;
[0012] All the aforementioned sensitive feature sets are then subjected to normalization and clustering processes in sequence to obtain multiple clusters.
[0013] The multiple clusters are reconstructed to obtain the self-noise separation result of the unmanned underwater vehicle;
[0014] The number of signal modal components is greater than the number of self-noise signal sources.
[0015] In an exemplary embodiment of this application, the self-noise signal source includes a narrowband line spectrum component, a broadband continuous spectrum component, and a modulation component mixed together;
[0016] The narrowband line spectrum component is generated by periodic mechanical noise and propeller noise; the broadband continuous spectrum component is generated by flow noise; and the sources of the modulation components include mechanical vibration, operation of electronic devices, water flow, and underwater sound wave reflection.
[0017] In an exemplary embodiment of this application, the expression for modal decomposition into a plurality of signal mode components is as follows:
[0018]
[0019] Where x(t) represents the noise signal source, f n (t) represents the nth signal mode component, N represents the number of signal mode components, n = 1, 2, ..., N, r n-1 (t) represents the remainder after modal decomposition of the (n-1)th signal modal component.
[0020] In an exemplary embodiment of this application, the time-domain feature set includes at least a first time-domain feature, a second time-domain feature, and a third time-domain feature;
[0021] The expression for the first time-domain feature is:
[0022]
[0023] Where t1 represents the first time-domain feature, x(m) represents the m-th time-domain signal, and M represents the number of time-domain signals, m = 1, 2, ..., M;
[0024] The expression for the second time-domain feature is:
[0025]
[0026] Where t2 represents the second time-domain feature;
[0027] The expression for the third time-domain feature is:
[0028]
[0029] Where t3 represents the third time-domain feature.
[0030] In an exemplary embodiment of this application, the frequency domain feature set includes a first frequency domain feature, a second frequency domain feature, a third frequency domain feature, a fourth frequency domain feature, a fifth frequency domain feature, and a sixth frequency domain feature;
[0031] The expression for the first frequency domain feature is:
[0032]
[0033] Where f1 represents the first frequency domain feature, s(k) represents the spectral value of the k-th frequency domain signal, and K represents the number of frequency domain signals, k = 1, 2, ..., K;
[0034] The expression for the second frequency domain feature is:
[0035]
[0036] Where f2 represents the second frequency domain feature;
[0037] The expression for the third frequency domain feature is:
[0038]
[0039] Where f3 represents the third frequency domain feature;
[0040] The expression for the fourth frequency domain feature is:
[0041]
[0042] Where f4 represents the fourth frequency domain feature;
[0043] The expression for the fifth frequency domain feature is:
[0044]
[0045] Where f5 represents the fifth frequency domain feature, Q k This represents the frequency of the k-th frequency domain signal;
[0046] The expression for the sixth frequency domain feature is:
[0047]
[0048] Where f6 represents the sixth frequency domain feature, s(h) represents the spectral value of the h-th spectral line, H represents the number of spectral lines, h = 1, 2, ..., H, Q' h This represents the frequency of the h-th spectral line;
[0049] The expression for the comprehensive feature set is:
[0050]
[0051] in, Represents the comprehensive feature set, t a Representing time-domain features, f b It represents the frequency domain characteristics.
[0052] In an exemplary embodiment of this application, the step of filtering the comprehensive feature set using feature distance evaluation technology to obtain multiple sensitive feature sets includes:
[0053] Multiple pattern classes are divided from the comprehensive feature set. Each pattern class contains multiple features, and each feature corresponds to a feature vector. All the pattern classes form a joint feature vector set.
[0054] Calculate the average intra-class distance between all feature vectors of each pattern class, and calculate the intra-class deviation factor for each average intra-class distance.
[0055] Calculate the inter-class distances between all the pattern classes, and calculate the inter-class deviation factor for each inter-class distance;
[0056] Based on all the inter-class deviation factors and all the intra-class deviation factors, multiple compensation factors are calculated respectively;
[0057] Using all the inter-class distances, all the average intra-class distances, and all the compensation factors, multiple feature distance evaluation metrics are calculated;
[0058] According to the order of all the feature distance evaluation metrics from largest to smallest, extract the corresponding number of features from each of the pattern classes to form multiple sensitive feature sets;
[0059] Wherein, the i-th sensitive feature set is represented as: S sen (i).
[0060] In an exemplary embodiment of this application, the expression for the joint feature vector set is:
[0061] {P (w,q) ,w=1,2,...,c; q=1,2,...j,...,g,...,N o} (12)
[0062] Among them, P (w,q) Let N represent the feature corresponding to the q-th feature vector in the w-th pattern class, where c represents the number of pattern classes. o Indicates the number of eigenvectors;
[0063] The expression for the set of average intra-class distances is:
[0064]
[0065] Among them, S 类内 Let S be the set of average intra-class distances. w This represents the average intra-class distance of the w-th pattern class. g represents the g-th eigenvector, j represents the j-th eigenvector, and P (w,g) P represents the feature corresponding to the g-th feature vector in the w-th pattern class. (w,j) This represents the feature corresponding to the j-th feature vector in the w-th pattern class;
[0066] The expression for the intra-class deviation factor is:
[0067]
[0068] Among them, V 类内 Represents the within-class deviation factor, max(S) w ) represents the maximum average intra-class distance, min(S) w ) represents the minimum average intra-class distance;
[0069] The expression for the set of inter-class distances is:
[0070]
[0071] Among them, S 类间 μ represents the inter-class distance. (w) Let represent the mean of all feature vectors in the w-th pattern class. μ' represents the mean of all pattern classes.
[0072] The expression for the inter-class deviation factor is:
[0073]
[0074] Among them, V 类内 Indicates the inter-class deviation factor;
[0075] The expression for the compensation factor is:
[0076]
[0077] Where λ represents the compensation factor;
[0078] The expression for the feature distance evaluation index is:
[0079]
[0080] Among them, J A This represents the feature distance evaluation metric.
[0081] In an exemplary embodiment of this application, the normalization process is performed on all the sensitive feature sets respectively to obtain a plurality of normalized feature sets, and each of the normalized feature sets includes a plurality of normalized features;
[0082] The expression for the normalized feature is:
[0083]
[0084] in, P represents the normalized feature corresponding to the φ-th feature vector in the i-th sensitive feature set. (i,φ) This represents the feature corresponding to the φ-th feature vector in the i-th sensitive feature set;
[0085] The normalized feature set is represented as:
[0086] In an exemplary embodiment of this application, the step of performing the clustering process on all the normalized feature sets respectively to obtain a plurality of clusters includes:
[0087] Multiple cluster centers are randomly selected from each of the normalized feature sets, and initialization processing is performed on all the cluster centers in each of the normalized feature sets.
[0088] Each normalized feature set is assigned a feature vector cluster that is closest to all cluster centers to obtain a feature vector cluster corresponding to each normalized feature set.
[0089] Calculate the average value of all feature vectors in each feature vector cluster, and correct all cluster centers in each normalized feature set to obtain all updated feature vectors in each normalized feature set;
[0090] Calculate the average value of all updated feature vectors in each of the normalized feature sets;
[0091] The algorithm iteratively repeats the above allocation, calculation, and update steps until the average value of all the feature vectors in each feature vector cluster no longer changes, at which point the algorithm converges and multiple clusters are obtained.
[0092] In an exemplary embodiment of this application, the step of reconstructing the plurality of clusters to obtain the self-noise separation result of the unmanned underwater vehicle includes:
[0093] Each of the clusters is assigned to a corresponding signal mode component, and the number of clusters is equal to the number of self-noise signal sources.
[0094] The signal modal components corresponding to each cluster are respectively divided into different categories of self-noise components;
[0095] The self-noise separation result of the unmanned underwater vehicle is obtained by summing all the signal mode components of the self-noise components of different categories.
[0096] Beneficial effects:
[0097] This application provides a method for separating UUV self-noise through decomposition, classification, and reconstruction, which has at least the following features:
[0098] Beneficial effects:
[0099] (1) This application decomposes the self-noise signal source of the unmanned underwater vehicle into multiple signal mode components; extracts features from all signal mode components to obtain a comprehensive feature set; then uses the feature distance evaluation technique to select multiple sensitive feature sets from the comprehensive feature set, and performs normalization and clustering processing on all sensitive feature sets in sequence to obtain multiple clusters; finally, by reconstructing multiple clusters, the self-noise of the unmanned underwater vehicle is effectively separated.
[0100] (2) At the same time, it achieves the effect of suppressing or offsetting the impact of the self-noise of the unmanned underwater vehicle on the detection performance. Attached Figure Description
[0101] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.
[0102] Figure 1This illustration shows a step diagram of a UUV self-noise separation method for decomposition, classification and reconstruction in an exemplary embodiment of this application;
[0103] Figure 2 This diagram illustrates a flowchart of a UUV self-noise separation method based on decomposition, classification, and reconstruction in an exemplary embodiment of this application.
[0104] Figure 3a A schematic diagram showing the time-domain waveform of the mixed self-noise composed of mechanical noise and flow noise in a simulation experiment of an exemplary embodiment of this application;
[0105] Figure 3b A schematic diagram showing the spectrum of the mixed self-noise composed of mechanical noise and flow noise in a simulation experiment of an exemplary embodiment of this application;
[0106] Figure 4 A schematic diagram showing the decomposition results of empirical mode decomposition in a simulation experiment of an exemplary embodiment of this application;
[0107] Figure 5a A schematic diagram showing the mechanical noise separation results obtained using the method proposed in this application in a simulation experiment of an exemplary embodiment of this application;
[0108] Figure 5b This diagram illustrates the flow noise separation results obtained using the method proposed in this application in a simulation experiment of an exemplary embodiment of this application. Detailed Implementation
[0109] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
[0110] Furthermore, the accompanying drawings are merely illustrative of this application and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.
[0111] This example implementation provides a UUV self-noise separation method based on decomposition, classification, and reconstruction (DCR), such as... Figure 1 As shown, the following steps may be included:
[0112] Step S101: Obtain the self-noise signal source of the unmanned underwater vehicle, and use the empirical mode decomposition algorithm to perform mode decomposition on the self-noise signal source to obtain multiple signal mode components.
[0113] Step S102: Extract the time-domain and frequency-domain features of multiple signal mode components respectively, construct the time-domain feature set and the frequency-domain feature set, and combine the time-domain feature set and the frequency-domain feature set into a comprehensive feature set.
[0114] Step S103: Use the feature distance evaluation technique to filter the comprehensive feature set to obtain multiple sensitive feature sets.
[0115] Step S104: Normalize and cluster all sensitive feature sets in sequence to obtain multiple clusters.
[0116] Step S105: Reconstruct the multiple clusters respectively to obtain the self-noise separation results of the unmanned underwater vehicle.
[0117] Among them, the number of signal modal components is greater than the number of self-noise signal sources.
[0118] This application proposes a method for separating UUV self-noise through decomposition, classification, and reconstruction, which has at least the following beneficial effects:
[0119] (1) This application decomposes the self-noise signal source of the unmanned underwater vehicle into multiple signal mode components; extracts features from all signal mode components to obtain a comprehensive feature set; then uses the feature distance evaluation technique to select multiple sensitive feature sets from the comprehensive feature set, and performs normalization and clustering processing on all sensitive feature sets in sequence to obtain multiple clusters; finally, by reconstructing multiple clusters, the self-noise of the unmanned underwater vehicle is effectively separated.
[0120] (2) At the same time, it achieves the effect of suppressing or offsetting the impact of the self-noise of the unmanned underwater vehicle on the detection performance.
[0121] The UUV self-noise separation method proposed in this example embodiment will be described in more detail below.
[0122] Self-noise separation of unmanned underwater vehicles (UUVs) under underdetermined conditions involves estimating a large number of quantities with a small number of quantities. Compared to positive definite and super-positive definite blind source separation, which has more observation signals, underdetermined blind source separation represents a fundamental change in processing methods. For underdetermined blind source separation, there is limited information available for direct use, necessitating the full exploration and utilization of the different latent characteristics of UUV self-noise.
[0123] Based on this, this application proposes a UUV self-noise separation method based on decomposition, classification, and reconstruction. In underdetermined conditions, especially when separating single-channel received signals, the core idea is as follows: First, the observed signal is decomposed into as many signal mode components as possible with different noise characteristics; then, features are extracted from each signal mode component to construct a comprehensive feature set; next, sensitive feature sets are selected from the comprehensive feature set using methods such as feature distance evaluation; third, the feature vectors of the sensitive feature set are classified into sets of self-noise signal sources using a clustering algorithm; finally, based on the classification results, multiple components are assigned to self-noise components from different sources, completing noise separation under underdetermined conditions.
[0124] Depend on Figure 2 As can be seen, the main components of the separation mode proposed in this application include: noise data decomposition, feature extraction, feature screening and classification, and signal reconstruction. Each part undertakes a different task and can be replaced by different algorithms and features.
[0125] Noise data decomposition essentially involves expanding the dimension of the self-noise signal source, converting a single-channel received signal into a multi-channel signal. This can be achieved using algorithms such as Empirical Mode Decomposition (EMD) or wavelet transform. Feature extraction is a prerequisite for effectively classifying the decomposition results. Feature selection and classification can be performed using clustering algorithms such as K-means or spectral clustering from machine learning. After signal reconstruction, the self-noise separation results of the unmanned underwater vehicle are obtained.
[0126] In step S101 of this embodiment, the EMD algorithm is preferably used to perform mode decomposition on the self-noise signal source to obtain multiple signal mode components.
[0127] The EMD algorithm is a method suitable for analyzing and processing nonlinear, non-stationary random signals, and has important applications in fields such as seismic signal processing and ocean wave signal processing. The EMD algorithm can decompose any signal into several Instrinsic Mode Function (IMF) components and a remainder term. The IMF component is a function or signal that satisfies the following two conditions:
[0128] The first condition is that the number of extreme points and the number of zero-crossing points in the entire data sequence must be equal or differ by at most one.
[0129] The second condition is that at any point, the mean of the upper envelope determined by the local maxima and the lower envelope determined by the local minima of the data sequence is zero, meaning the signal is locally symmetrical about the time axis.
[0130] The modal decomposition process of EMD is also a screening process, specifically:
[0131] Let the local mean of the self-noise signal source x(t) be m. 11 (t), x(t) and m 11 Let the difference between (t) be h. 11 (t), i.e., h 11 (t)=x(t)-m 11 (t). If h 11 (t) If the above IMF component conditions are not met, then repeat the above steps k times, i.e., execute h. 1k (t)=h 1(k-1) (t)-m 1k (t), until h 11 f1(t) satisfies the two conditions of the above IMF component, denoted as: f1(t) = h 1k (t).
[0132] Then, subtract f1(t) from x(t) to obtain the modal decomposition remainder r1(t) for the first decomposition, i.e., r1(t) = x(t) - f1(t). This remainder is then used as the signal yet to be decomposed, and the decomposition steps are repeated to sequentially decompose the modal decomposition remainders to obtain: The process continues until the information contained in the last modal decomposition remainder is of little significance to the research content, becoming a monotonic function from which IMF components can no longer be selected.
[0133] Thus, the self-noise signal source x(t) is decomposed into N signal mode components f. n (t) and a mode decomposition remainder r n-1 (t). The expression for the mode decomposition of multiple signal mode components is:
[0134]
[0135] Where x(t) represents the noise signal source, f n (t) represents the nth signal mode component, N represents the number of signal mode components, n = 1, 2, ..., N, r n-1(t) represents the remainder after modal decomposition of the (n-1)th signal modal component. These N signal modal components are arranged in descending order of frequency, with f1(t) containing the highest frequency components. n (t) contains the lowest frequency components. Modal decomposition remainder r n-1 (t) is a non-oscillating monotonic sequence.
[0136] It should also be noted that, in this embodiment, the self-noise signal source includes a mixture of narrowband line spectrum components, broadband continuous spectrum components, and modulation components. Specifically, periodic mechanical noise and propeller noise generate the narrowband line spectrum component; flow noise generates the broadband continuous spectrum component; and the sources of the modulation component include mechanical vibration, the operation of electronic devices, water flow, and acoustic wave reflections from the underwater environment.
[0137] In step S102 of this embodiment, time-domain features and frequency-domain features of multiple signal mode components are extracted respectively to construct time-domain feature sets and frequency-domain feature sets, and the time-domain feature sets and frequency-domain feature sets are combined into a comprehensive feature set.
[0138] In this embodiment, a time-domain feature set including a first time-domain feature, a second time-domain feature, and a third time-domain feature is constructed. Figure 2 In this context, S1 represents the time-domain feature set.
[0139] The expression for this first time-domain feature is:
[0140]
[0141] Where t1 represents the first time-domain feature, x(m) represents the m-th time-domain signal, M represents the number of time-domain signals, and m = 1, 2, ..., M.
[0142] The expression for this second time-domain feature is:
[0143]
[0144] Where t2 represents the second time-domain feature.
[0145] The expression for this third time-domain feature is:
[0146]
[0147] Where t3 represents the third time-domain feature.
[0148] In this embodiment, a frequency domain feature set including six frequency domain features was constructed. Figure 2 In this context, S2 represents the frequency domain feature set, and these six frequency domain features are as follows:
[0149] The expression for the first frequency domain feature is:
[0150]
[0151] Where f1 represents the first frequency domain feature, s(k) represents the spectral value of the k-th frequency domain signal, and K represents the number of frequency domain signals, k = 1, 2, ..., K.
[0152] The expression for the second frequency domain feature is:
[0153]
[0154] Where f2 represents the second frequency domain feature.
[0155] The expression for the third frequency domain feature is:
[0156]
[0157] Here, f3 represents the third frequency domain feature.
[0158] The expression for the fourth frequency domain feature is:
[0159]
[0160] Here, f4 represents the fourth frequency domain feature.
[0161] The expression for the fifth frequency domain feature is:
[0162]
[0163] Where f5 represents the fifth frequency domain feature, Q k This represents the frequency of the k-th frequency domain signal.
[0164] The expression for the sixth frequency domain feature is:
[0165]
[0166] Where f6 represents the sixth frequency domain feature, s(h) represents the spectral value of the h-th spectral line, H represents the number of spectral lines, h = 1, 2, ..., H, Q' h This represents the frequency of the h-th spectral line.
[0167] Finally, we obtain the representation of the comprehensive feature set as follows:
[0168]
[0169] in, Represents the comprehensive feature set, t a Representing time-domain features, f b It represents the frequency domain characteristics.
[0170] In step S103 of this embodiment, the comprehensive feature set is filtered using feature distance evaluation technology to obtain multiple sensitive feature sets. Step S103 may include the following sub-steps:
[0171] Sub-step S1031: Divide the comprehensive feature set into multiple pattern classes. Each pattern class contains multiple features, and each feature corresponds to a feature vector. All pattern classes form a joint feature vector set.
[0172] Furthermore, the expression for this joint feature vector set is:
[0173] {P (w,q) ,w=1,2,...,c; q=1,2,...j,...,g,...,N o} (12)
[0174] Among them, P (w,q) Let N represent the feature corresponding to the q-th feature vector in the w-th pattern class, where c represents the number of pattern classes. o This indicates the number of eigenvectors.
[0175] Sub-step S1032: Calculate the average intra-class distance between all feature vectors of each pattern class, and calculate the intra-class bias factor for each average intra-class distance.
[0176] Furthermore, the expression for the set of average intra-class distances is:
[0177]
[0178] Among them, S 类内 Let S be the set of average intra-class distances. w This represents the average intra-class distance of the w-th pattern class. g represents the g-th eigenvector, j represents the j-th eigenvector, and P (w,g) P represents the feature corresponding to the g-th feature vector in the w-th pattern class. (w,j) This represents the feature corresponding to the j-th feature vector in the w-th pattern class.
[0179] Furthermore, the expression for this intra-class deviation factor is:
[0180]
[0181] Among them, V 类内 Represents the within-class deviation factor, max(S) w ) represents the maximum average intra-class distance, min(S) w ) represents the minimum average intra-class distance.
[0182] Sub-step S1033: Calculate the inter-class distance between all pattern classes, and calculate the inter-class deviation factor for each inter-class distance.
[0183] Furthermore, the expression for the set of distances between these classes is:
[0184]
[0185] Among them, S 类间 μ represents the inter-class distance. (w) Let represent the mean of all feature vectors in the w-th pattern class. μ' represents the mean of all pattern classes.
[0186] Furthermore, the expression for the inter-class deviation factor is:
[0187]
[0188] Among them, V 类内 This represents the inter-class deviation factor.
[0189] Sub-step S1034: Calculate multiple compensation factors based on all inter-class deviation factors and all intra-class deviation factors.
[0190] Furthermore, the expression for this compensation factor is:
[0191]
[0192] Where λ represents the compensation factor.
[0193] Sub-step S1035: Calculate multiple feature distance evaluation metrics using all inter-class distances, all average intra-class distances, and all compensation factors.
[0194] Furthermore, the expression for this feature distance evaluation index is:
[0195]
[0196] Among them, J A This represents the feature distance evaluation metric.
[0197] Sub-step S1036: Extract the corresponding number of features from each pattern class according to the order of all feature distance evaluation metrics from largest to smallest, and form multiple sensitive feature sets.
[0198] The sensitive feature set is represented as: S sen (i), where i represents the i-th sensitive feature set.
[0199] Sort all feature distance evaluation metrics from largest to smallest to facilitate clustering later.
[0200] After obtaining the comprehensive feature set, directly classifying it would result in irrelevant or redundant features. Therefore, an effective feature selection method is needed. In this embodiment, a feature distance evaluation technique is preferably used to effectively select these comprehensive features, ultimately constructing a sensitive feature set for determining multiple signal mode components.
[0201] Based on the above feature distance evaluation method, it can be intuitively concluded that small average intra-class distance and large average inter-class distance have good separability in the feature domain. The distance evaluation index J obtained according to formula (18) is... A In descending order, a corresponding number of features are extracted from each pattern class to form multiple sensitive feature sets. In this embodiment, the corresponding number ranges from 3 to 8.
[0202] In step S104 of this embodiment, normalization and clustering are performed sequentially on all obtained sensitive datasets to obtain multiple clusters. Step S104 of this embodiment includes the following sub-steps:
[0203] Sub-step S1041: Normalize all sensitive feature sets separately to obtain multiple normalized feature sets, each of which includes multiple normalized features. The expression for each normalized feature is:
[0204]
[0205] in, P represents the normalized feature corresponding to the φ-th feature vector in the i-th sensitive feature set. (i,φ) This represents the feature corresponding to the φ-th feature vector in the i-th sensitive feature set.
[0206] The normalized feature set is represented as:
[0207] Sub-step S1042: Perform clustering on all normalized feature sets to obtain multiple clusters.
[0208] After obtaining the normalized feature set, clustering is performed on it. Commonly used clustering methods include K-means clustering, manual clustering, and kernel density estimation. Among these, K-means clustering is the most commonly used method in cluster analysis due to its simplicity, convenience, and fast convergence. This embodiment uses the K-means clustering method to cluster the normalized feature set.
[0209] The specific process is as follows:
[0210] The first step is to randomly select multiple cluster centers from each normalized feature set, and then initialize all cluster centers in each normalized feature set.
[0211] After initialization, the initial average value of the feature vectors and the initial objective function can be obtained:
[0212] The expression for the initial average value is:
[0213]
[0214] Where J represents the initial average value of the feature vectors, L represents the number of sensitive feature sets, i = 1, 2, ..., L, J i This represents the initial objective function.
[0215] The expression for the initial objective function is:
[0216]
[0217] Where ||·|| represents the linear distance between two vector matrices in space, D e Let P represent the e-th cluster center. (e,s) Let z represent the feature corresponding to the s-th feature vector that is closest to the e-th cluster center, where e = 1, 2, ..., z, and z represents the number of cluster centers.
[0218] The second step is to assign the feature vectors that are closest to all cluster centers in each normalized feature set to obtain the feature vector clusters corresponding to each normalized feature set.
[0219] The expression for this feature vector cluster is:
[0220] D e ={P (e,s) |e=arg min||P (e,s) -D e 2 ||} (22)
[0221] Here, argmin represents the parameter value when the function takes its minimum value in its domain.
[0222] The third step is to calculate the average value of all feature vectors in each feature vector cluster, and then correct all cluster centers in each normalized feature set to obtain all updated feature vectors in each normalized feature set.
[0223] The expression for correcting the cluster centers of the feature vector clusters is:
[0224] D e ←E[P (e,s) ]P∈z (twenty three)
[0225] Where E represents the expected value, and ← represents the cluster center adjustment based on the expected value.
[0226] The fourth step is to calculate the average value of all updated feature vectors in each normalized feature set; then, a second correction is made to all cluster centers in each normalized feature set.
[0227] The fifth step involves iteratively repeating the above allocation, calculation, and update steps until the average value of all eigenvectors in the eigenvector cluster no longer changes, at which point the algorithm converges and multiple clusters are obtained.
[0228] Step S105 in this embodiment may include the following sub-steps:
[0229] Sub-step S1051: Assign each cluster to a signal mode component, and make the number of clusters equal to the number of self-noise signal sources.
[0230] Sub-step S1052: Divide the signal mode components corresponding to each cluster into different categories of self-noise components.
[0231] Sub-step S1053: Sum all signal mode components in different categories of self-noise components to obtain the self-noise separation result of the unmanned underwater vehicle.
[0232] To verify the superior performance of the UUV self-noise separation method proposed in this application, which involves decomposition, classification, and reconstruction, the following simulation experiments were conducted.
[0233] When faced with numerous separation algorithms to choose from, it is crucial to effectively evaluate their performance. Establishing an objective separation evaluation system is an essential step in describing algorithmic separation performance. Before using the aforementioned EMD algorithm for UUV self-noise separation, it is necessary to analyze the performance evaluation metrics of each algorithm. Signal-based evaluation metrics assess the similarity or closeness between the source signal and the corresponding separated signal, primarily including correlation coefficient, mean square error (MSE), and the improvement in signal-to-noise ratio (SNR).
[0234] If the correlation coefficient is 1, it indicates that the estimated signal and the noise signal source are consistent; if it is 0, it indicates that they are completely inconsistent. In the correlation coefficient matrix formed by all the results, if only one number in each row approaches 1 and the others approach 0, it indicates that the overall separation effect of the algorithm is good.
[0235] The smaller the mean square error, the closer the noise source and the estimated signal are, and the better the performance of the separation algorithm.
[0236] This simulation experiment employed both mixed-type (two types) and mixed-type (three types) UUV simulated self-noise. The mixing ratio of the mixed-type (two types) UUV simulated self-noise was 1:1, and the mixing ratio of the mixed-type (three types) UUV simulated self-noise was 1:1:1. The number of received signal channels for the simulated self-noise was one.
[0237] like Figure 3a , Figure 3b and Figure 4 As shown, the simulation of UUV self-noise involves two types: mechanical noise and flow noise. First, the EMD algorithm is used to decompose the mixed mechanical and flow noise into 15 IMF components. These IMF components are then sorted according to their energy values from largest to smallest, and the top 10 IMF components by energy value are retained. This process is then further refined. Figure 3a The time-domain waveforms of the first 10 IMF components are shown. This flow noise is preferably propeller noise. Figure 3b The results show that the majority of IMF components with line spectral elements are IMF1 to IMF5, while IMF6 to IMF10 are mostly continuous spectral components. See Table 1 below:
[0238] Table 1: IMF Component Classification Table
[0239]
[0240] As can be seen from Table 1, IMF1 to IMF5 belong to mechanical noise, while IMF6 to IMF10 belong to flow noise.
[0241] Using formulas (2) to (10), features are extracted from mechanical noise, flow noise, and all retained IMF components, resulting in three feature vector sets: a mechanical noise feature vector set, a flow noise feature vector set, and an IMF component feature vector set. Using feature distance evaluation techniques, feature vectors of mechanical noise and flow noise are filtered to obtain sensitive feature sets for these two types of self-noise, as shown in Table 2 below.
[0242] Table 2: Feature Types of Sensitive Features
[0243]
[0244] Table 2 shows that the time-domain features in this sensitive feature set include: t imf3 t imf6 and t imf9 Frequency domain characteristics include: f imf3 f imf4 f imf5 and fimf6 .
[0245] t imf3 Representing the time-domain characteristics of IMF3, t imf6 Representing the time-domain characteristics of IMF6, t imf9 This represents the time-domain characteristics of IMF9.
[0246] f imf3 The frequency domain characteristics of IMF3 are represented by f. imf4 This represents the frequency domain characteristics of IMF4, f imf5 This represents the frequency domain characteristics of IMF5, f imf6 This represents the frequency domain characteristics of IMF6.
[0247] like Figure 5a and Figure 5b As shown, the DCR separation mode proposed in this application can basically recover mechanical noise and flow noise with good spectral characteristics. Figure 5a As can be seen, the line spectrum component of the estimated mechanical noise is well recovered, while the continuous spectrum component is significantly lost because most of the energy of the mechanical noise is concentrated in the line spectrum component. Figure 5b As can be seen, the estimated flow noise spectrum was recovered very well.
[0248] The self-noise of mixed mechanical noise-flow noise, mechanical noise-propeller noise, flow noise-propeller noise, and mechanical noise-propeller noise-flow noise was further separated using the methods described above. The separation results for each mixed mode are shown in Table 3 below:
[0249] Table 3: Separation results under various mixing modes using DCR separation mode
[0250]
[0251] Table 3 shows the correlation coefficient, mean square error, and signal-to-noise ratio (SNR) for each hybrid mode. In the mode of mixing two types of UUV self-noise, the DCR separation mode has the best separation effect when separating the mixed mechanical noise and first-order noise; however, it is the weakest when separating the mixed flow noise and propeller noise. Overall, the DCR separation mode achieves similar results in separating various hybrid self-noise types. However, when separating the self-noise of the mixed mechanical noise and propeller noise first-order noise, all three evaluation indicators decrease, indicating that the DCR separation mode is significantly weaker in separating three types of hybrid self-noise than in separating two types. This is because the overlap of noise frequency bands is more severe when mixing three types of self-noise, thus reducing the separation effect.
[0252] Simulation results show that when two types of self-noise are mixed, the average correlation coefficient of the DCR-separated signal is 0.81, and the average signal-to-noise ratio is improved by 5.17 dB. When three types of self-noise are mixed, the average correlation coefficient of the DCR-separated signal is 0.72, and the average signal-to-noise ratio is improved by 4.44 dB. This demonstrates that the UUV self-noise separation method proposed in this application, involving decomposition, classification, and reconstruction, achieves good noise separation performance.
[0253] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly specified.
[0254] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.
[0255] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the scope of the technology disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application.
[0256] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.
Claims
1. A method for separating UUV self-noise through decomposition, classification, and reconstruction, characterized in that, Includes the following steps: The self-noise signal source of the unmanned underwater vehicle is acquired, and the self-noise signal source is decomposed into multiple signal mode components using the empirical mode decomposition algorithm. The time-domain and frequency-domain features of multiple signal mode components are extracted respectively to construct a time-domain feature set and a frequency-domain feature set, and the time-domain feature set and the frequency-domain feature set are combined into a comprehensive feature set; The comprehensive feature set is filtered using a feature distance evaluation technique to obtain multiple sensitive feature sets, including: Multiple pattern classes are divided from the comprehensive feature set. Each pattern class contains multiple features, and each feature corresponds to a feature vector. All the pattern classes form a joint feature vector set. Calculate the average intra-class distance between all feature vectors of each pattern class, and calculate the intra-class deviation factor for each average intra-class distance. Calculate the inter-class distances between all the pattern classes, and calculate the inter-class deviation factor for each inter-class distance; Based on all the inter-class deviation factors and all the intra-class deviation factors, multiple compensation factors are calculated respectively; Using all the inter-class distances, all the average intra-class distances, and all the compensation factors, multiple feature distance evaluation metrics are calculated; According to the order of all the feature distance evaluation metrics from largest to smallest, extract the corresponding number of features from each of the pattern classes to form multiple sensitive feature sets; Wherein, the i-th sensitive feature set is represented as: S sen (i); All the aforementioned sensitive feature sets are then subjected to normalization and clustering processes in sequence to obtain multiple clusters. The multiple clusters are reconstructed to obtain the self-noise separation result of the unmanned underwater vehicle; The number of signal modal components is greater than the number of self-noise signal sources.
2. The UUV self-noise separation method based on decomposition, classification, and reconstruction according to claim 1, characterized in that, The self-noise signal source includes narrowband line spectrum components, broadband continuous spectrum components, and modulation components mixed together; The narrowband line spectrum component is generated by periodic mechanical noise and propeller noise; the broadband continuous spectrum component is generated by flow noise; and the sources of the modulation components include mechanical vibration, operation of electronic devices, water flow, and underwater sound wave reflection.
3. The UUV self-noise separation method based on decomposition, classification, and reconstruction according to claim 1, characterized in that, The expression for modal decomposition into multiple signal modal components is as follows: Where x(t) represents the noise signal source, f n (t) represents the nth signal mode component, N represents the number of signal mode components, n = 1, 2, ..., N, r n-1 (t) represents the remainder after modal decomposition of the (n-1)th signal modal component.
4. The UUV self-noise separation method based on decomposition, classification, and reconstruction according to claim 3, characterized in that, The time-domain feature set includes at least a first time-domain feature, a second time-domain feature, and a third time-domain feature; The expression for the first time-domain feature is: Where t1 represents the first time-domain feature, x(m) represents the m-th time-domain signal, and M represents the number of time-domain signals, m = 1, 2, ..., M; The expression for the second time-domain feature is: Where t2 represents the second time-domain feature; The expression for the third time-domain feature is: Where t3 represents the third time-domain feature.
5. The UUV self-noise separation method based on decomposition, classification, and reconstruction according to claim 4, characterized in that, The frequency domain feature set includes a first frequency domain feature, a second frequency domain feature, a third frequency domain feature, a fourth frequency domain feature, a fifth frequency domain feature, and a sixth frequency domain feature; The expression for the first frequency domain feature is: Where f1 represents the first frequency domain feature, s(k) represents the spectral value of the k-th frequency domain signal, and K represents the number of frequency domain signals, k = 1, 2, ..., K; The expression for the second frequency domain feature is: Where f2 represents the second frequency domain feature; The expression for the third frequency domain feature is: Where f3 represents the third frequency domain feature; The expression for the fourth frequency domain feature is: Where f4 represents the fourth frequency domain feature; The expression for the fifth frequency domain feature is: Where f5 represents the fifth frequency domain feature, Q k This represents the frequency of the k-th frequency domain signal; The expression for the sixth frequency domain feature is: Where f6 represents the sixth frequency domain feature, s(h) represents the spectral value of the h-th spectral line, H represents the number of spectral lines, h = 1, 2, ..., H, Q' h This represents the frequency of the h-th spectral line; The expression for the comprehensive feature set is: in, Represents the comprehensive feature set, t a Representing time-domain features, f b It represents the frequency domain characteristics.
6. The UUV self-noise separation method based on decomposition, classification, and reconstruction according to claim 5, characterized in that, The expression for the joint feature vector set is: {P (w,q) ,w=1,2,...,c;q=1,2,...j,...,g,...,N o } (12) Among them, P (w,q) Let N represent the feature corresponding to the q-th feature vector in the w-th pattern class, where c represents the number of pattern classes. o Indicates the number of eigenvectors; The expression for the set of average intra-class distances is: Among them, S 类内 Let S be the set of average intra-class distances. w This represents the average intra-class distance of the w-th pattern class. g represents the g-th eigenvector, j represents the j-th eigenvector, and P (w,g) P represents the feature corresponding to the g-th feature vector in the w-th pattern class. (w,j) This represents the feature corresponding to the j-th feature vector in the w-th pattern class; The expression for the intra-class deviation factor is: Among them, V 类内 Represents the within-class deviation factor, max(S) w ) represents the maximum average intra-class distance, min(S) w ) represents the minimum average intra-class distance; The expression for the set of inter-class distances is: Among them, S 类间 μ represents the inter-class distance. (w) Let represent the mean of all feature vectors in the w-th pattern class. μ' represents the mean of all pattern classes. The expression for the inter-class deviation factor is: Among them, V 类内 Indicates the inter-class deviation factor; The expression for the compensation factor is: Where λ represents the compensation factor; The expression for the feature distance evaluation index is: Among them, J A This represents the feature distance evaluation metric.
7. The UUV self-noise separation method based on decomposition, classification, and reconstruction according to claim 6, characterized in that, The normalization process is performed on all the sensitive feature sets respectively to obtain multiple normalized feature sets, each of which includes multiple normalized features; The expression for the normalized feature is: in, P represents the normalized feature corresponding to the φ-th feature vector in the i-th sensitive feature set. (i,φ) This represents the feature corresponding to the φ-th feature vector in the i-th sensitive feature set; The normalized feature set is represented as:
8. The UUV self-noise separation method based on decomposition, classification, and reconstruction according to claim 7, characterized in that, The step of performing the clustering process on all the normalized feature sets to obtain multiple clusters includes: Multiple cluster centers are randomly selected from each of the normalized feature sets, and initialization processing is performed on all the cluster centers in each of the normalized feature sets. Each normalized feature set is assigned a feature vector cluster that is closest to all cluster centers to obtain a feature vector cluster corresponding to each normalized feature set. Calculate the average value of all feature vectors in each feature vector cluster, and correct all cluster centers in each normalized feature set to obtain all updated feature vectors in each normalized feature set; Calculate the average value of all updated feature vectors in each of the normalized feature sets; The algorithm iteratively repeats the above allocation, calculation, and update steps until the average value of all the feature vectors in each feature vector cluster no longer changes, at which point the algorithm converges and multiple clusters are obtained.
9. The UUV self-noise separation method based on decomposition, classification, and reconstruction according to claim 8, characterized in that, The step of reconstructing the multiple clusters to obtain the self-noise separation result of the unmanned underwater vehicle includes: Each of the clusters is assigned to a corresponding signal mode component, and the number of clusters is equal to the number of self-noise signal sources. The signal modal components corresponding to each cluster are respectively divided into different categories of self-noise components; The self-noise separation result of the unmanned underwater vehicle is obtained by summing all the signal mode components of the self-noise components of different categories.