CN102692625A - A Feature Joint Modeling Method of Bottom Target Echo and Reverberation in Rn Space - Google Patents

Abstract

What the present invention provides is a feature joint modeling method of underwater target echo and reverberation in a kind of R ⁿ space, and its steps are: 1. Establish the training sample storehouse that comprises target echo and reverberation, according to two kinds of signal physics Due to the difference in nature, different signal processing methods are used to map the training samples to multiple signal feature spaces; 2. According to the distribution information of the training samples in the signal feature space, the signal feature space is reduced in dimension, and different signal feature spaces Fusion, establish the fusion signal feature space; 3. Establish the joint distribution model of the target echo and reverberation in the training sample in the fusion signal feature space; 4. For unknown signal samples, pass the signal processing described in steps 1 and 2 The method is mapped to the fusion feature space, and compared with the joint distribution model established in step 3, to determine the type of unknown signal samples. The established feature space proposed by the present invention can make the target echo and reverberation more stable and universally separable.

Description

A Feature Joint Modeling Method of Bottom Target Echo and Reverberation in Rn Space

technical field

The invention relates to the application field of underwater acoustic technology, in particular to a characteristic joint modeling method of underwater target echo and reverberation in ^Rn space.

Background technique

The detection of underwater sunken or buried targets plays an important role in underwater tasks such as underwater archaeology and shipwreck salvage. According to the detection distance, underwater target detection can be divided into short-distance detection and long-distance detection. In short-distance detection, imaging sonar and other equipment can be used to obtain acoustic images of the target and its vicinity, and intuitively identify and judge the target. In the case of long-distance detection, the target does not have the imaging conditions, and the shape information of the target cannot be obtained. The existence of the target can only be judged according to the vibration information of the target modulated in the echo signal.

At present, when detecting long-distance underwater targets, the usual method is to use active sonar to transmit a broadband pulse, and use a sonar array to receive the echo signal. According to the difference in physical mechanism between target echo and reverberation, determine the difference in signal properties between the two, and select the corresponding mathematical transformation method to extract the signal characteristics of the echo signal, and use the pattern recognition method to analyze the signal of the echo signal. Features are judged and identified. In this signal processing flow, the current research and achievements are mainly focused on analyzing the properties of the target echo and reverberation, and studying the corresponding signal feature extraction methods. In this direction, domestic scholars have proposed a target echo bright spot model suitable for engineering applications, defined the basic forms of geometric echo and elastic echo in target echo, and adopted a variety of methods based on time domain and frequency domain. And signal processing methods in the time-frequency domain to extract signal features. However, the methods that have been studied so far are to suppress reverberation as an interference signal when extracting signal features, and there are no relevant literature reports on the signal properties and characteristics of reverberation.

Contents of the invention

The purpose of the present invention is to propose a feature of underwater target echo and reverberation in R ⁿ space that has optimal separability in the sense of Euclidean distance, can improve the detection performance of active sonar, and realize remote safe detection. joint modeling approach.

The purpose of the present invention is achieved like this:

A feature joint modeling method of underwater target echo and reverberation in R ⁿ space comprises the following steps:

Step 1: Taking reverberation as a signal source with stable signal properties, establish a training sample library containing target echo and reverberation, and use different signal processing methods to map the training samples to multiple in a signal feature space;

Step 2: According to the distribution information of the training samples in the signal feature space, reduce the dimensionality of the signal feature space, and fuse different signal feature spaces to establish a fusion signal feature space;

Step 3: Establish the joint distribution model of target echo and reverberation in the fusion signal feature space in the training samples;

Step 4: For unknown signal samples, map them into the fusion feature space through the signal processing methods described in steps 1 and 2, and compare with the joint distribution model established in step 3 to determine the type of unknown signal samples.

The use of different signal processing methods refers to the use of fractional Fourier transform to map the training samples to the energy-aggregating feature space and the use of Hilbert-Huang transform to map the training samples to the multi-component feature space.

The dimensionality reduction of the signal feature space is performed by linear discriminant analysis based on the Fisher criterion function in the sense of Euclidean distance.

The signal feature space fusion is carried out by using the method of canonical correlation analysis combined with serial feature fusion.

The joint distribution modeling of the feature space adopts a discriminant function method to establish a mathematical equation on the classification surface of the target echo and reverberation in the fusion feature space.

Main features of the method of the present invention:

The present invention regards the reverberation as a signal source with stable signal properties, and establishes a fusion feature space R ⁿ according to the difference in signal properties between the target echo and the reverberation, combined with a feature space compression and fusion method. Compared with the existing method for establishing signal feature space, the feature space established by the method of the present invention can make the target echo and reverberation more stable and universally separable. The achievements of the present invention are not limited to the application of underwater target recognition, but can also be widely used in active sonar target recognition fields such as forward-looking sonar, underwater unmanned submersible, frogman sonar, etc., to improve the detection of active sonar performance, to achieve long-range security detection, and to provide a new solution for offshore defense sonar.

Description of drawings

Fig. 1 is a schematic diagram of implementation steps of the joint modeling and identification method of the present invention.

Detailed ways

The present invention is described in more detail below in conjunction with accompanying drawing example:

Combined with Figure 1. Taking Fractional Fourier Transform and Hilbert-Huang Transform as examples, the content described here does not limit the content of the present invention. The present invention comprises the following steps:

step 1:

According to the bright spot model of the target echo signal, when the active sonar emission signal is a chirp signal, the target echo is also a chirp signal, which has the characteristics of energy aggregation and multi-component. According to this feature, the fractional Fourier transform and Hilbert-Huang transform can be used to extract the features of the echo signal, and the energy aggregation feature and multi-component feature of the echo signal can be extracted respectively.

For the echo signal x(t), the fractional Fourier transform shown in formula (1) is performed on it.

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{<span\; class="notranslate">\; -</span><span\; class="notranslate">\; \&infin;</span>}{\overset{<span\; class="notranslate">\; +</span><span\; class="notranslate">\; \&infin;</span>}{<span\; class="notranslate">\; \&Integral;</span>}}{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; dt</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

Among them, the kernel function K _p (u, t) is a function related to time t, the coordinate axis u of the fractional-order domain, and the fractional-order power p. By changing the value of p, the peak value of f _p (u) is maximized, at this time f _p (u) is the energy aggregation characteristic of the echo signal.

The Hilbert-Huang transformation of the echo signal x(t) is to perform EMD decomposition on x(t), fit the envelope of the signal, establish the establishment condition of the IMF component, and subtract the average value of the upper and lower envelopes from the signal The IMF components of each order are obtained step by step in the same way, and finally the signal is decomposed into the sum of several IMF components and a remainder. Abandoning the remaining items and performing Hilbert transformation on the remaining IMF components can obtain the Hilbert spectrum of the echo signal.

Through the above two signal feature extraction methods, the two feature spaces in step 1 in Fig. 1 are established.

Step 2: Contains two steps of dimensionality reduction and fusion of signal feature space.

Through step 1, the echo signal x(t) is mapped to a specific signal feature space from the time domain. To compress this signal feature space, one principle that should be followed is that in the compressed signal feature space, the inter-class separability of target echo and reverberation is higher. Therefore, in the sense of Euclidean distance, the Fisher criterion function shown in formula (2) is used as the compression criterion of the signal feature space. The purpose is that in the compressed signal feature space, the two types of signal samples have a smaller intra-class Scatter distance, with larger inter-class scatter distances.

${\mathrm{<span\; class="notranslate">\; J</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; w</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; b</span>}}\mathrm{<span\; class="notranslate">\; w</span>}}{{\mathrm{<span\; class="notranslate">\; w</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; w</span>}}\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

Among them, _JF is the objective function reflecting the separability between two types of signal classes, w is the projection direction vector used to project the signal feature space to reduce the dimensionality, S _w is the intra-class scatter matrix of signal samples in the feature space, S _b is the inter-class scatter matrix of signal samples in feature space. Formula (2) is solved by Lagrange multiplier method, and a set of projection direction vectors satisfying the orthogonal condition are obtained.

在这组矢量上进行投影，就得到了新的特征空间，定义为R′_i，如附图中步骤2所示。R′_i的空间维度要低于原信号特征空间

这样就实现了信号特征空间的降维。The signal feature space established in step 1

Projecting on this group of vectors, a new feature space is obtained, which is defined as R′ _i , as shown in step 2 in the attached figure. The spatial dimension of R′ _i is lower than the original signal feature space

In this way, the dimensionality reduction of the signal feature space is realized.

分别采用上述的特征压缩方法得到两个低维特征空间R′₁和R′₂。为了将这两个信号特征空间的信息进行融合，采用典型相关分析提取两个信号特征空间中代表同一个信号样本的特征矢量之间的典型相关变量。将两个信号特征空间中代表同一个信号样本的特征矢量的典型相关变量进行首尾相连的拼接，就得到一个新的融合特征空间，定义为Rⁿ。在本例中，n＝2d。For the two signal feature spaces obtained through step 1 and

Two low-dimensional feature spaces R' ₁ and R' ₂ are obtained by using the above-mentioned feature compression method respectively. In order to fuse the information of the two signal feature spaces, canonical correlation analysis is used to extract the canonical correlation variables between the feature vectors representing the same signal sample in the two signal feature spaces. A new fused feature space is obtained by concatenating the canonical correlation variables representing the feature vectors of the same signal sample in the two signal feature spaces end to end, which is defined as R ⁿ . In this example, n=2d.

Step 3:

In the fusion feature space R ⁿ , according to the distribution of the target echo and reverberation in the feature space, a classification surface g(z) described by a mathematical equation can be established between the two types of signal samples, and the decision based on this classification surface The rules state that,

The form of g(z) is a linear weighted combination of each dimension of the feature space, and each weight value can be obtained by an optimization method based on known samples. Let g(z) be zero, and what is obtained is the mathematical equation of the classification surface between the two types of signals in the feature space. According to this equation, the distribution model of the target echo and reverberation in the feature space ^Rn can be established.

Step 4:

For the newly obtained echo signal sample, use the method described in step 1 and step 2 to extract its signal feature, and obtain its feature vector z in the fusion feature space, and substitute it into the decision rule. According to the decision rule, you can Determine the category of z.

Claims (5)

1. a feature joint modeling method of underwater target echo and reverberation in R ⁿ space, it is characterized in that comprising the following steps: Step 1: Taking reverberation as a signal source with stable signal properties, establish a training sample library containing target echo and reverberation, and use different signal processing methods to map the training samples to multiple in a signal feature space; Step 2: According to the distribution information of the training samples in the signal feature space, reduce the dimension of the signal feature space, and fuse different signal feature spaces to establish a fusion signal feature space; Step 3: Establish the joint distribution model of target echo and reverberation in the fusion signal feature space in the training samples; Step 4: For unknown signal samples, map them into the fusion feature space through the signal processing methods described in steps 1 and 2, and compare them with the joint distribution model established in step 3 to determine the type of unknown signal samples. 2. the feature joint modeling method of the underwater target echo and reverberation in a kind of R ⁿ space according to claim 1, it is characterized in that: described adopting different signal processing methods refers to adopting fractional order Fourier The leaf transformation maps the training samples to the energy-aggregating feature space and uses the Hilbert-Huang transform to map the training samples to the multi-component feature space. 3. the feature joint modeling method of the underwater target echo and reverberation in a kind of R ⁿ space according to claim 1, it is characterized in that: described signal characteristic space dimensionality reduction, what adopted is in Euclidean distance In the sense, linear discriminant analysis based on Fisher's criterion function was performed. 4. the feature joint modeling method of the underwater target echo and reverberation in a kind of R ⁿ space according to claim 1, it is characterized in that: described signal characteristic space fusion, what adopted is typical correlation analysis in conjunction with series connection The method of feature fusion is carried out. 5. the feature joint modeling method of the underwater target echo and reverberation in a kind of R ⁿ space according to claim 1, it is characterized in that: described feature space joint distribution modeling adopts discriminant function method A mathematical equation is established for the classification surface of the target echo and reverberation in the fusion feature space.

Images