CN114280566B - One-dimensional range profile identification method based on class label association - Google Patents

Abstract

The invention discloses a method for identifying a one-dimensional range profile associated with a class of tags, and belongs to the technical field of radar target identification. The class label association one-dimensional range profile recognition method utilizes the training data set and the corresponding class label vector to construct the class label association subspace, and the class information is directly obtained by carrying out association projection on the one-dimensional range profile sample of the input target, so that a classifier is not required to be introduced to carry out final class decision, the problem that the recognition performance is reduced due to the adoption of the classifier in the conventional recognition method is avoided, and the recognition rate of the target is improved.

Description

One-dimensional range image recognition method based on class label association

technical field

The invention belongs to the technical field of radar target recognition, and in particular relates to a one-dimensional range image recognition method associated with class tags.

Background technique

The one-dimensional range image obtained by broadband radar contains the information of the target structure and shape, which is very beneficial to target classification, so that higher classification performance can be obtained. Compared with the two-dimensional radar image, it is easier to obtain and harder to identify One-dimensional range image is more advantageous, therefore, one-dimensional range image recognition has become a hot spot in the field of radar target recognition.

The conventional radar target recognition method still belongs to the classic pattern recognition method. First, the classification features of the target must be extracted, and then the classifier is used to make the final decision on the target category. However, the classifier may introduce certain classification errors, which will lead to the recognition of the entire system rate drops. Therefore, there is room for further improvement in the performance of conventional object recognition methods.

Contents of the invention

The invention provides a one-dimensional range image recognition method associated with class tags, which can be used to improve the performance of radar target recognition.

The technical scheme adopted in the present invention is:

A class label associated one-dimensional distance image recognition method, the method comprising:

Define x _ij to represent the j-th training one-dimensional range image of the i-th known target, 1≤i≤g, 1≤j≤N _i , where g is the number of categories, N _i is the i-th known target number of training samples;

Define y _ij to represent a class label vector with a one-dimensional distance like x _ij , where y _ij is a g-dimensional column vector, the value of the i-th element is 1, and the value of other elements is 0;

Calculate the autocorrelation matrix R _XY = XY ^T of one-dimensional range image samples, where the one-dimensional range image sample matrix

Calculate the cross-correlation matrix R _XY =XY ^T between one-dimensional range image samples and corresponding class label vectors, where the class label matrix

Construct Correlation Subspace Based on Autocorrelation Matrix R _XX and Cross Correlation Matrix R _XY

Transform the input one-dimensional distance image x _t of the target to be identified to obtain its class label vector y _t =W ^T x _t , where the vector y _t is a g-dimensional column vector;

Determine its category based on the class label vector y _t of the one-dimensional distance image x _t : traverse each element in the vector y _t , if the current element (y _t,k ) is smaller than the rest of the elements (ie y _t,k >y _{t, l} , l=1,2,...g,l≠k), then it is determined that the target category of the one-dimensional range image x _t is the category corresponding to the current element, that is, it belongs to the kth category.

In order to further ensure the recognition performance, the autocorrelation matrix R _XX between the one-dimensional range image sample and the corresponding class label vector is calculated according to the formula R _XX =(XX ^T +λI), where I represents the identity matrix and λ represents the adjustment factor.

The technical solution provided by the present invention brings at least the following beneficial effects:

The present invention uses the training data set and the corresponding class label vector to construct the class label associated subspace, and directly obtains the category information to which it belongs by performing associated projection on the one-dimensional range image sample of the input target, without introducing a classifier for the final category Decision-making avoids the decline in recognition performance of conventional recognition methods due to the use of classifiers, thereby improving the recognition rate of targets.

Detailed ways

In order to make the purpose, technical solution and advantages of the present invention clearer, the implementation manners of the present invention will be further described in detail below.

The embodiment of the present invention provides a class label associated target recognition method, which projects the one-dimensional range image sample of the input target through the associated transformation subspace, and directly obtains the corresponding class label information, avoiding the identification caused by the introduction of the classifier Performance drops, resulting in improved target recognition.

An embodiment of the present invention provides a class label associated target recognition method, including:

Class label association identification method:

其中，g为类别数，N_i为第i类已知目标的训练样本数，N为训练样本总数。x_ij表示相应的类标签矢量为y_ij，是g维列矢量，第i个元素的值为1，其它元素的值为0。由x_ij组成一维距离像样本矩阵X：Let x _ij (n-dimensional column vector) be the j-th training one-dimensional range image of the i-th type of known target, 1≤i≤g, 1≤j≤N _i ,

Among them, g is the number of categories, N _i is the number of training samples of the i-th known target, and N is the total number of training samples. x _ij indicates that the corresponding class label vector is y _ij , which is a g-dimensional column vector, the value of the i-th element is 1, and the value of other elements is 0. A one-dimensional distance image sample matrix X is composed of x _ij :

The class label matrix Y is constructed from the class label vector y _ij :

Calculate the autocorrelation matrix R _XX of one-dimensional range image samples:

R _XX = XX ^T (3)

Among them, the superscript "T" means matrix transposition. In order to avoid R _XX being a singular matrix, an adjustment factor λ is introduced. λ is an empirical value, and the specific value can be determined by experiments.

R _XX ＝(XX ^T +λI) (4)

Among them, I is the identity matrix. Calculate the cross-correlation matrix R _XY between one-dimensional range image samples and corresponding class label vectors

R _XY = XY ^T (5)

Based on the sample autocorrelation matrix and the sample-class label cross-correlation matrix, the associated subspace W is constructed:

Perform the following transformation on the input one-dimensional distance image x _t of the target to be identified to obtain the corresponding class label vector y _t :

y _t = W ^T x _t (7)

If satisfied:

where y _t,k and y _t,l are the elements in the vector y _t . Then it is judged that x _t belongs to the kth class.

In order to verify the recognition performance of a kind of tag-associated one-dimensional distance image recognition method provided by the embodiment of the present invention, the following simulation experiment is carried out:

Set four point targets: '|', 'V', 'Dry' and 'Small' targets. The bandwidth of the radar emission pulse is 150MHZ (distance resolution is 1m, radar radial sampling interval is 0.5m), the target is set as a uniform scattering point target, the scattering points of the "|" target are 5, and the scattering points of the other three targets are 9. In the one-dimensional range images with target attitude angles ranging from 0° to 70° at intervals of 1°, take the one-dimensional distances with target attitude angles of 0°, 2°, 4°, 6°,...,70° The image is trained, and the one-dimensional range image of the other attitude angles is used as the test data, so there are 35 test samples for each type of target.

For four kinds of targets ("|" font target, "V" font target, "dry" font target and "small" font target), within the range of attitude angle 0°～70°, using the embodiment of the present invention The provided class label association recognition method and the conventional recognition method based on feature subspace have carried out recognition experiments, and the average recognition rate of 4 types of targets reaches 87%, while the average recognition rate of the conventional recognition method based on feature subspace is 82%. %, thus verifying that the recognition method provided by the embodiment of the present invention can indeed improve the recognition performance of multi-type targets.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still be Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent replacements are made to some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

What have been described above are only some embodiments of the present invention. For those skilled in the art, without departing from the inventive concept of the present invention, several modifications and improvements can be made, and these all belong to the protection scope of the present invention.

Claims (2)

1. A method for class tag-associated one-dimensional distance image recognition, characterized in that, comprising the following steps: Define x _ij to represent the j-th training one-dimensional range image of the i-th known target, 1≤i≤g, 1≤j≤N _i , where g is the number of target categories, and N _i is the i-th known target The number of training samples; Define y _ij to represent a class label vector with a one-dimensional distance like x _ij , where y _ij is a g-dimensional column vector, the value of the i-th element of the class label vector is 1, and the value of other elements is 0; Calculate the autocorrelation matrix R _XX of the one-dimensional range image sample = (XX ^T + λI), wherein, the one-dimensional range image sample matrix

I represents the identity matrix, and λ represents the adjustment factor; Calculate the cross-correlation matrix R _XY =XY ^T between one-dimensional range image samples and corresponding class label vectors, where the class label matrix

Construct Correlation Subspace Based on Autocorrelation Matrix R _XX and Cross Correlation Matrix R _XY

Transform the input one-dimensional range image x _t of the target to be identified to obtain its class label vector y _t = W ^T x _t ; Determine its category based on the class label vector y _t of the one-dimensional distance image x _t : Traverse each element in the vector y _t , compare the current element with the rest of the vector y _t , if they are less than, then the current element corresponds to The category of is the target category of the one-dimensional distance image x _t . 2. The method according to claim 1, characterized in that, calculate the autocorrelation matrix R _XX between the one-dimensional range image sample and the corresponding class label vector according to the formula R XX =(XX ^T +λI ₎ , wherein, I represents The identity matrix, λ represents the preset adjustment factor.