One-dimensional range profile recognition method based on bidirectional time sequence convolution

Abstract

The embodiment of the invention relates to the technical field of radar target recognition, in particular to a one-dimensional range profile recognition method based on bidirectional time sequence convolution, and the method comprises the steps: carrying out the preprocessing of obtained sample one-dimensional range profile data, marking a real label, and carrying out the division according to a preset proportion, thereby obtaining a training set and a test set; constructing a one-dimensional range profile recognition model, wherein the one-dimensional range profile recognition model is composed of a bidirectional time sequence convolutional network serving as a feature extractor and a classifier; performing supervised iterative training on the one-dimensional range profile recognition model based on the training set to continuously realize network parameter tuning of the one-dimensional range profile recognition model until the one-dimensional range profile recognition model converges, and obtaining an optimal model; and inputting the one-dimensional range profile data of the moving target to be identified into the optimal model to obtain an identification result of the one-dimensional range profile data of the moving target output by the optimal model. According to the method, the precision and robustness of moving target classification can be well improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of radar target recognition technology, and in particular to a one-dimensional range profile recognition method based on bidirectional temporal convolution. Background Art

[0002] Mobile target type identification is an important part of the intelligence command system. Accurate target type identification helps to accurately estimate threats and reasonably allocate firepower, thereby maximizing combat effectiveness. Mobile target type identification can be divided into one-dimensional range images, SAR (Synthetic Aperture Radar) images, infrared data, optical images, and multi-source data fusion according to the type of data. Different types of data have different characteristics and applicable scenarios. Technicians can choose the optimal strategy based on target characteristics and application scenarios.

[0003] HRRP (High Resolution Range Profile) is the target echo signal received by broadband radar. It consists of multiple independent range units along the radar line of sight. Each range unit is composed of scattering points of different intensities, reflecting the sum of the echoes of important components of the target. One-dimensional range profile data provides rich descriptive information for target classification and identification by reflecting a series of important features such as the target's geometric structure, scattering intensity distribution, and physical size. It is an important data basis for radar target identification.

[0004] As the echo data of broadband radar, one-dimensional range profile has the characteristics of amplitude sensitivity, translation sensitivity and attitude sensitivity. Amplitude sensitivity reflects that the reflection intensity of the target is affected by the material and surface characteristics, translation sensitivity causes the position change of the target in the direction of the radar line of sight, causing feature displacement, and attitude sensitivity is manifested as the significant impact of different angles or directions of the target on the range profile characteristics. Although these characteristics provide rich target information, they may have an adverse effect on feature extraction and classification results, thereby limiting classification accuracy and classification stability.

[0005] Commonly used one-dimensional range image recognition methods are divided into two categories: traditional algorithms and deep learning algorithms. Traditional algorithms include statistical feature extraction, template matching, dimensionality reduction, and transformation. The one-dimensional range image classification algorithm based on deep learning automatically extracts the separable features of one-dimensional range image data through end-to-end supervised learning.

[0006] Traditional algorithms usually rely on artificially designed features and find it difficult to fully capture the complexity of the target. For example, statistical feature extraction methods may ignore key details for targets with complex shapes and find it difficult to effectively capture local characteristics, resulting in reduced classification performance; template matching methods are highly dependent on the position of the scattering center, and changes in the position or posture of the target will significantly reduce the matching degree between the template and the actual range image; dimensionality reduction methods may lose some important information during the dimensionality reduction process and find it difficult to reflect the timing or position correlation information between different scattering points. Compared with traditional algorithms, deep learning algorithms can automatically extract global and local complex features of the target through end-to-end learning, significantly reducing reliance on artificially designed features. However, deep learning algorithms are not sufficiently adaptable to the complexity of one-dimensional range image data, resulting in limited robustness to translation and posture changes. If preprocessing is insufficient, it may be difficult to extract translation-invariant features or cross-posture features. Summary of the invention

[0007] In view of this, an embodiment of the present application proposes a one-dimensional range profile recognition method based on bidirectional temporal convolution, which can improve the accuracy and robustness of mobile target classification, and effectively solve the problem that the traditional one-dimensional range profile recognition method is easily affected by azimuth sensitivity, resulting in low classification accuracy and poor stability.

[0008] In a first aspect, an embodiment of the present application proposes a one-dimensional range image recognition method based on bidirectional temporal convolution, the method comprising: preprocessing the acquired sample one-dimensional range image data, and after marking the real label, dividing it into a training set and a test set according to a preset ratio; constructing a one-dimensional range image recognition model, the one-dimensional range image recognition model consisting of a bidirectional temporal convolution network as a feature extractor and a classifier; wherein the bidirectional temporal convolution network is used to perform forward temporal processing and backward temporal processing on the input one-dimensional range image data, establish a bidirectional temporal relationship between different distance units of the one-dimensional range image, so as to extract the translation invariant feature, and the classifier is used to perform category recognition based on the translation invariant feature and output the recognition result; based on the training set, supervised iterative training is performed on the one-dimensional range image recognition model to continuously optimize the network parameters of the one-dimensional range image recognition model until the one-dimensional range image recognition model converges and obtains the optimal one-dimensional range image recognition model; the one-dimensional range image data of the moving target to be identified is input into the optimal one-dimensional range image recognition model, and the recognition result of the one-dimensional range image data of the moving target output by the optimal one-dimensional range image recognition model is obtained.

[0009] Optionally, the acquired sample one-dimensional range image data is preprocessed, including: denoising the acquired sample one-dimensional range image data, smoothing the noise in the sample one-dimensional range image data, and improving the separability between different categories; intercepting the denoised sample one-dimensional range image data with a preset length based on the center point of the target, and if the length is insufficient, zero-filling is performed to make up the preset length, so as to obtain the preprocessed sample one-dimensional range image data.

[0010] Optionally, the bidirectional temporal convolutional network adopts a Bi-LSTM network structure, which is specifically composed of a forward LSTM layer, a backward LSTM layer and an output layer. The forward LSTM layer and the backward LSTM layer both contain multiple LSTM feature learning units. The forward LSTM layer and the backward LSTM layer respectively obtain the temporal information in the input one-dimensional distance image from two different time directions; the forward LSTM layer is used to capture the context information from the starting point to the current distance unit to obtain the forward hidden state; the backward LSTM layer is used to capture the context information from the current distance unit to the end point to obtain the backward hidden state; the output layer is used to splice the forward hidden state and the backward hidden state to finally obtain the translation invariant feature.

[0011] Optionally, the translation invariant features finally extracted by the bidirectional temporal convolutional network are expressed by the formula:

[0012] H n =[h1,h2,…,h n ,…,h N ];

[0013]

[0014]

[0015] Among them, H n represents the translation invariant features finally extracted by the bidirectional temporal convolutional network, Cat(·) represents the splicing of the channel dimension, and are the forward hidden state and backward hidden state of the nth distance unit, LSTM forward (·) represents the forward LSTM layer, LSTM backward (·) represents the backward LSTM layer, N represents the total number of range units, and prof(n) represents the echo strength of the nth range unit.

[0016] Optionally, the LSTM feature learning unit is composed of an input gate, a forget gate, an output gate, and a memory unit. The input and output functions of the input gate, the forget gate, and the output gate are expressed by the formula:

[0017] i n =σ[Whi h n-1 +W xi prof(n)+b i ];

[0018] f n =σ[W hf h n-1 +W xf prof(n)+b f ];

[0019] y n =σ[W hy h n-1 +W xy prof(n)+b y ];

[0020] Among them, prof(n) is the input of the input gate, forget gate and output gate, i n represents the output of the input gate, f n represents the output of the forget gate, y n represents the output of the output gate, σ represents the Sigmoid activation function, W xi , W xf and W xy is the weight matrix of the input gate, forget gate and output gate for the current input, W hi , W hf and W hy is the weight matrix of the input gate, forget gate and output gate for the previous input, b i 、b f and b y Bias for input gate, forget gate and output gate;

[0021] Suppose the value stored in the memory cell at the current moment is C n , C n The calculation formula is:

[0022] C n =C n-1 ×f n +tanh(W c [h n-1 ]+b c )×prof(n);

[0023] Among them, C m-1 represents the value stored in the memory cell at the previous moment, tanh(n) represents the Tanh activation function, W c is the weight matrix of the memory unit for the current input, b c is the bias of the memory unit;

[0024] The output of the LSTM feature learning unit at the current moment is expressed by the formula:

[0025]

[0026] in, Represents the output of the LSTM feature learning unit at the current moment.

[0027] Optionally, the classifier is composed of a first fully connected layer, a ReLU activation layer, a second fully connected layer and a Softmax layer connected sequentially. The classifier performs category recognition based on translation invariant features and outputs recognition results, which is expressed by the formula:

[0028]

[0029] f classifer (H n ) = f softmax {f Linear2 [f ReLU (f Linear1 [H n ])]};

[0030] Among them, f classifer (·) represents the classifier, f Linear1 (·) represents the first fully connected layer, f ReLU (·) represents the ReLU activation layer, f Linear2 (·) represents the second fully connected layer, f softmax represents the Softmax layer, Represents the output recognition result.

[0031] Optionally, supervised iterative training is performed on the one-dimensional range image recognition model based on the training set to continuously optimize the network parameters of the one-dimensional range image recognition model until the one-dimensional range image recognition model converges to obtain an optimal one-dimensional range image recognition model, including: inputting the sample one-dimensional range image data in the training set into the one-dimensional range image recognition model in batches for supervised training to obtain a recognition result of the sample one-dimensional range image data output by the one-dimensional range image recognition model; using a cross entropy loss function to calculate a loss value between the true label of the sample one-dimensional range image data and the recognition result of the sample one-dimensional range image data output by the one-dimensional range image recognition model; based on the loss value, using a stochastic gradient descent method to continuously optimize and adjust the network parameters of the one-dimensional range image recognition model until the one-dimensional range image recognition model converges to obtain an optimal one-dimensional range image recognition model.

[0032] Optionally, a cross entropy loss function is used to calculate the loss value between the true label of the sample one-dimensional range image data and the recognition result of the sample one-dimensional range image data output by the one-dimensional range image recognition model, which is expressed by the formula:

[0033]

[0034] Where B represents the batch size, D represents the total number of categories, and CE(·) represents the cross entropy loss function. represents the true label of the one-dimensional distance image data of the b-th sample, It represents the recognition result of the one-dimensional range image data of the b-th sample output by the one-dimensional range image recognition model, and LOSS represents the calculated loss value.

[0035] The one-dimensional range profile recognition method based on bidirectional temporal convolution proposed in this application has the following positive effects.

[0036] First, after collecting the sample one-dimensional range image data, the present application needs to perform preprocessing to reduce the interference of signal noise, and construct a training set based on the preprocessed sample one-dimensional range image data. Using such a training set to train the one-dimensional range image recognition model can enable the one-dimensional range image recognition model to fully exploit the translation-invariant features of the moving target, thereby effectively improving the recognition accuracy of the one-dimensional range image recognition model.

[0037] Second, the one-dimensional range image recognition model constructed in this application uses a bidirectional temporal convolutional network as a feature extractor. The bidirectional temporal convolutional network can capture the global dependency of the one-dimensional range image data according to the forward and backward information flows, thereby enhancing the robustness of the one-dimensional range image recognition model to the translation and posture changes of moving targets, thereby effectively improving the recognition performance, stability and reliability of the one-dimensional range image recognition model.

[0038] Third, the one-dimensional range image recognition model constructed in this application is reasonably designed and has a simple structure. It does not need to rely on artificially designed features, can make full use of the local and global characteristics of the one-dimensional range image data, has the ability to quickly extract features and classify targets, and has achieved high-precision and high-efficiency recognition of mobile targets. It has strong adaptability and can meet the complex needs of mobile target recognition, and is of great value to both military and civilian fields.

[0039] In the second aspect, an embodiment of the present application proposes a one-dimensional range image recognition system based on bidirectional temporal convolution, the system comprising: a sample division module, a model construction module, a model training module and a model use module; the sample division module is used to pre-process the acquired sample one-dimensional range image data, and after marking the real label, divide it into a training set and a test set according to a preset ratio; the model construction module is used to construct a one-dimensional range image recognition model, the one-dimensional range image recognition model is composed of a bidirectional temporal convolution network as a feature extractor and a classifier; wherein the bidirectional temporal convolution network is used to perform forward temporal processing and backward temporal processing on the input one-dimensional range image data, and establish The bidirectional temporal relationship between different distance units of the one-dimensional range image is used to extract the translation invariant features. The classifier is used to perform category recognition based on the translation invariant features and output the recognition results; the model training module is used to perform supervised iterative training on the one-dimensional range image recognition model based on the training set to continuously optimize the network parameters of the one-dimensional range image recognition model until the one-dimensional range image recognition model converges and obtains the optimal one-dimensional range image recognition model; the model use module is used to input the one-dimensional range image data of the moving target to be identified into the optimal one-dimensional range image recognition model to obtain the recognition result of the one-dimensional range image data of the moving target output by the optimal one-dimensional range image recognition model.

[0040] It can be understood that the beneficial effects of the second aspect mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the related technologies, the drawings required for use in the embodiments of the present application or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 is a flowchart of a one-dimensional range profile recognition method based on bidirectional temporal convolution provided in one embodiment of the present application;

[0043] Figure 2 is a structural schematic diagram of a one-dimensional range profile recognition model provided in an embodiment of the present application;

[0044] Figure 3 is a structural schematic diagram of a one-dimensional range profile recognition system based on bidirectional temporal convolution provided in an embodiment of the present application;

[0045] Figure 4 It is a structural schematic diagram of an electronic device provided in another embodiment of the present application. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below in conjunction with the accompanying drawings. In the various embodiments of the present application, many technical details are proposed in order to make the reader better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can also be implemented. The division of the following embodiments is only for the convenience of description, and the specific implementation mode of the present application should not constitute any limitation. The various embodiments can be combined with each other and referenced to each other under the premise of no contradiction.

[0047] An embodiment of the present application proposes a one-dimensional range profile recognition method based on bidirectional temporal convolution, which is applied to an electronic device, wherein the electronic device may be a terminal or a server. In this embodiment and the following embodiments, the electronic device is described using a server as an example. The implementation details of the one-dimensional range profile recognition method based on bidirectional temporal convolution proposed in this embodiment are specifically described below. The following content is only the implementation details provided for ease of understanding and is not necessary for the implementation of this solution.

[0048] The specific process of the one-dimensional range profile recognition method based on bidirectional temporal convolution proposed in this embodiment can be as follows: Figure 1 As shown, including:

[0049] Step 101 , preprocessing the acquired sample one-dimensional range image data, and after marking the real label, dividing it into a training set and a test set according to a preset ratio.

[0050] In the specific implementation, the server first needs to obtain several sample one-dimensional range image data of the sample target, and then preprocess the obtained sample one-dimensional range image data to remove noise and normalize it. After completing the preprocessing, the sample one-dimensional range image data is labeled with real labels. After all the sample one-dimensional range image data are labeled, they are divided into training set and test set according to the preset ratio.

[0051] In one example, when the server preprocesses the acquired sample one-dimensional range image data, it first performs denoising on the acquired sample one-dimensional range image data, that is, smoothes the noise in the sample one-dimensional range image data to improve the separability between different categories. After completing the denoising operation, the server needs to cut the denoised sample one-dimensional range image data to a preset length based on the center point of the target. If the denoised sample one-dimensional range image data is not long enough, it needs to be zero-filled to fill the preset length, and finally obtain the preprocessed sample one-dimensional range image data.

[0052] It can be understood that the preprocessing operation can make the sample one-dimensional range image data in the training set and the test set cleaner and more reliable. Using such a training set to train the one-dimensional range image recognition model can improve the training quality of the one-dimensional range image recognition model and enable the one-dimensional range image recognition model to quickly acquire recognition capabilities.

[0053] Step 102, construct a one-dimensional range image recognition model, which is composed of a bidirectional temporal convolutional network as a feature extractor and a classifier, wherein the bidirectional temporal convolutional network is used to perform forward temporal processing and backward temporal processing on the input one-dimensional range image data, establish a bidirectional temporal relationship between different distance units of the one-dimensional range image, so as to extract translation-invariant features, and the classifier is used to perform category recognition based on the translation-invariant features and output the recognition results.

[0054] In the specific implementation, after the server completes the construction of the training set and the test set, it can build a one-dimensional range image recognition model. The one-dimensional range image recognition model is specifically composed of a bidirectional temporal convolutional network as a feature extractor and a classifier. The bidirectional temporal convolutional network is used to perform forward and backward temporal processing on the input one-dimensional range image data, establish a bidirectional temporal relationship between different distance units of the one-dimensional range image, and extract translation-invariant features. The classifier is used to perform category recognition based on the translation-invariant features and output the recognition results.

[0055] In an example, the specific structure of the one-dimensional range profile recognition model constructed by the server can be as follows: Figure 2 shown.

[0056] In one example, the bidirectional temporal convolutional network adopts the Bi-LSTM network structure, which is specifically composed of a forward LSTM layer, a backward LSTM layer and an output layer. The forward LSTM layer and the backward LSTM layer both contain multiple LSTM feature learning units. The forward LSTM layer and the backward LSTM layer respectively obtain the temporal information in the input one-dimensional range image from two different time directions, and the output layer can merge the temporal information in the two directions at each time step of the one-dimensional range image data. The forward LSTM layer is specifically used to capture the context information from the starting point to the current distance unit to obtain the forward hidden state, the backward LSTM layer is specifically used to capture the context information from the current distance unit to the end point to obtain the backward hidden state, and the output layer is used to splice the forward hidden state and the backward hidden state to finally obtain the translation invariant feature.

[0057] In an example, the translation-invariant features finally extracted by the bidirectional temporal convolutional network are expressed as:

[0058] H n =[h1,h2,…,h n ,…,h N];

[0059]

[0060] Among them, H n represents the translation invariant features finally extracted by the bidirectional temporal convolutional network, Cat(·) represents the splicing of the channel dimension, and are the forward hidden state and backward hidden state of the nth distance unit, LSTM forward (·) represents the forward LSTM layer, LSTM backward (·) represents the backward LSTM layer, N represents the total number of range units, and prof(n) represents the echo strength of the nth range unit.

[0061] Understandably, H n It contains both the forward information from the starting point to the nth distance unit and the backward information from the nth distance unit to the end point, which is a comprehensive expression of bidirectional features. Establishing a bidirectional temporal relationship between different distance units of one-dimensional range image data can effectively alleviate the problem of posture sensitivity in one-dimensional range image data, thereby improving the recognition performance of the target corresponding to the one-dimensional range image data.

[0062] In an example, the LSTM feature learning unit consists of an input gate, a forget gate, an output gate, and a memory unit. The input and output functions of the input gate, the forget gate, and the output gate can be expressed by the formula:

[0063] i n =σ[W hi h n-1 +W xi prof(n)+b i ];

[0064] f n =σ[W hf h n-1 +W xf prof(n)+b f ];

[0065] y n =σ[W hy h n-1 +W xy prof(n)+b y ];

[0066] Among them, prof(n) is the input of the input gate, forget gate and output gate, i n represents the output of the input gate, f n represents the output of the forget gate, y n represents the output of the output gate, σ represents the Sigmoid activation function, Wxi , W xf and W xy is the weight matrix of the input gate, forget gate and output gate for the current input, W hi , W hf and W hy is the weight matrix of the input gate, forget gate and output gate for the previous input, b i 、b f and b y is the bias of the input gate, forget gate and output gate.

[0067] Suppose the value stored in the memory cell at the current moment is C n , C n The calculation formula can be expressed as:

[0068] C n =C n-1 ×f n +tanh(W c [h n-1 ]+b c )×prof(n);

[0069] Among them, C n-1 represents the value stored in the memory cell at the previous moment, tanh(n) represents the Tanh activation function, W c is the weight matrix of the memory unit for the current input, b c is the bias of the memory cell.

[0070] The output of the LSTM feature learning unit at the current moment can be expressed by the formula:

[0071]

[0072] in, Represents the output of the LSTM feature learning unit at the current moment.

[0073] In an example, the classifier is specifically composed of a first fully connected layer, a ReLU activation layer, a second fully connected layer, and a Softmax layer connected sequentially. The classifier performs category recognition based on translation-invariant features and outputs recognition results, which can be specifically expressed by the formula:

[0074]

[0075] f classifer (H n )=f softmax {f Linear2 [f ReLU (f Linear1 [H n ])]};

[0076] Among them, f classifer (·) represents the classifier, f Linear1 (·) represents the first fully connected layer, f ReLU (·) represents the ReLU activation layer, f Linear2 (·) represents the second fully connected layer, f softmax represents the Softmax layer, Represents the output recognition result.

[0077] Step 103 , supervised iterative training is performed on the one-dimensional range profile recognition model based on the training set to continuously optimize the network parameters of the one-dimensional range profile recognition model until the one-dimensional range profile recognition model converges to obtain the optimal one-dimensional range profile recognition model.

[0078] In a specific implementation, after the server completes the construction of the one-dimensional range image recognition model, it can perform supervised iterative training on the one-dimensional range image recognition model based on the training set to continuously optimize the network parameters of the one-dimensional range image recognition model until the one-dimensional range image recognition model converges and obtains the optimal one-dimensional range image recognition model.

[0079] In one example, the server needs to input the sample one-dimensional range image data in the training set into the one-dimensional range image recognition model in batches for supervised training, and obtain the recognition results of the sample one-dimensional range image data output by the one-dimensional range image recognition model. Then, the cross entropy loss function is used to calculate the loss value between the true label of the sample one-dimensional range image data and the recognition result of the sample one-dimensional range image data output by the one-dimensional range image recognition model. Finally, based on the loss value, the network parameters of the one-dimensional range image recognition model are continuously optimized and adjusted using the stochastic gradient descent method until the one-dimensional range image recognition model converges to obtain the optimal one-dimensional range image recognition model.

[0080] In one example, the server uses the cross entropy loss function to calculate the loss value between the true label of the sample one-dimensional range image data and the recognition result of the sample one-dimensional range image data output by the one-dimensional range image recognition model, which can be achieved by the following formula:

[0081]

[0082] Where B represents the batch size, D represents the total number of categories, and CE(·) represents the cross entropy loss function. represents the true label of the one-dimensional distance image data of the b-th sample, It represents the recognition result of the one-dimensional range image data of the b-th sample output by the one-dimensional range image recognition model, and LOSS represents the calculated loss value.

[0083] In one example, the test set is used to perform various performance tests on the optimal one-dimensional range image recognition model. The server inputs the sample one-dimensional range image data in the test set into the optimal one-dimensional range image recognition model, and obtains various test indicators based on the recognition results of the sample one-dimensional range image data in the test set output by the optimal one-dimensional range image recognition model, as well as the true labels of the sample one-dimensional range image data in the test set. If one or several tests fail, the server will return the optimal one-dimensional range image recognition model for retraining and network parameter tuning.

[0084] Step 104 , inputting the one-dimensional range profile data of the moving target to be identified into the optimal one-dimensional range profile recognition model, and obtaining the recognition result of the one-dimensional range profile data of the moving target output by the optimal one-dimensional range profile recognition model.

[0085] In a specific implementation, after obtaining the optimal one-dimensional range image recognition model, the server can deploy the optimal one-dimensional range image recognition model to an application scenario in need. When using the one-dimensional range image recognition model, the one-dimensional range image data of the moving target to be identified is input into the optimal one-dimensional range image recognition model to obtain the recognition result of the one-dimensional range image data of the moving target output by the optimal one-dimensional range image recognition model.

[0086] The one-dimensional range profile recognition method based on bidirectional temporal convolution proposed in this embodiment has the following positive effects.

[0087] First, after collecting the sample one-dimensional range image data, the present application needs to perform preprocessing to reduce the interference of signal noise, and construct a training set based on the preprocessed sample one-dimensional range image data. Using such a training set to train the one-dimensional range image recognition model can enable the one-dimensional range image recognition model to fully exploit the translation-invariant features of the moving target, thereby effectively improving the recognition accuracy of the one-dimensional range image recognition model.

[0088] Second, the one-dimensional range image recognition model constructed in this application uses a bidirectional temporal convolutional network as a feature extractor. The bidirectional temporal convolutional network can capture the global dependency of the one-dimensional range image data according to the forward and backward information flows, thereby enhancing the robustness of the one-dimensional range image recognition model to the translation and posture changes of moving targets, thereby effectively improving the recognition performance, stability and reliability of the one-dimensional range image recognition model.

[0089] Third, the one-dimensional range image recognition model constructed in this application is reasonably designed and has a simple structure. It does not need to rely on artificially designed features, can make full use of the local and global characteristics of the one-dimensional range image data, has the ability to quickly extract features and classify targets, and has achieved high-precision and high-efficiency recognition of mobile targets. It has strong adaptability and can meet the complex needs of mobile target recognition, and is of great value to both military and civilian fields.

[0090] The step division of the above methods is only for clear description. When implemented, they can be combined into one step or some steps can be split and decomposed into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this application; adding insignificant modifications to the algorithm or process or introducing insignificant designs without changing the core design of the algorithm and process are all within the scope of protection of this application.

[0091] Another embodiment of the present application proposes a one-dimensional range profile recognition system based on bidirectional temporal convolution. The details of the one-dimensional range profile recognition system based on bidirectional temporal convolution proposed in this embodiment are specifically described below. The following content is only the implementation details provided for the convenience of understanding and is not necessary for the implementation of this embodiment. Figure 3 This embodiment proposes a one-dimensional range profile recognition system based on bidirectional temporal convolution, and the system includes: a sample division module 201, a model construction module 202, a model training module 203 and a model use module 204.

[0092] The sample division module 201 is used to pre-process the acquired sample one-dimensional range image data, and after marking the real labels, divide them into a training set and a test set according to a preset ratio.

[0093] The model building module 202 is used to build a one-dimensional range image recognition model, which is composed of a bidirectional temporal convolutional network as a feature extractor and a classifier; wherein the bidirectional temporal convolutional network is used to perform forward temporal processing and backward temporal processing on the input one-dimensional range image data, establish a bidirectional temporal relationship between different distance units of the one-dimensional range image, so as to extract translation-invariant features, and the classifier is used to perform category recognition based on the translation-invariant features and output the recognition results.

[0094] The model training module 203 is used to perform supervised iterative training on the one-dimensional range profile recognition model based on the training set to continuously optimize the network parameters of the one-dimensional range profile recognition model until the one-dimensional range profile recognition model converges to obtain the optimal one-dimensional range profile recognition model.

[0095] The model using module 204 is used to input the one-dimensional range image data of the mobile target to be identified into the optimal one-dimensional range image recognition model to obtain the recognition result of the one-dimensional range image data of the mobile target output by the optimal one-dimensional range image recognition model.

[0096] It is worth mentioning that all modules involved in this embodiment are logic modules. In practical applications, a logic unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed by this application, but this does not mean that there are no other units in this embodiment.

[0097] It is not difficult to find that this embodiment is a system embodiment corresponding to the above method embodiment, and this embodiment can be implemented in conjunction with the above method embodiment. The relevant technical details and technical effects mentioned in the above embodiments are still valid in this embodiment, and in order to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied in the above embodiments.

[0098] Another embodiment of the present application deploys an electronic device, whose specific structure is as follows Figure 4 As shown, it includes: at least one processor 301; and a memory 302 that is communicatively connected to the at least one processor 301; wherein the memory 302 stores instructions that can be executed by the at least one processor 301, and the instructions are executed by the at least one processor 301 so that the at least one processor 301 can execute a one-dimensional range image recognition method based on bidirectional temporal convolution described in the above embodiments.

[0099] Among them, the memory and the processor can be connected in a bus manner, and the bus can include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and memories together. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are all well known in the art and will not be further described in this article. The bus interface is responsible for providing an interface between the bus and the transceiver. The transceiver can be one component or multiple components, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor is transmitted on a wireless medium through an antenna, and further, the antenna also receives data and transmits the data to the processor.

[0100] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory can be used to store data used by the processor when performing operations.

[0101] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a device (such as a single-chip microcomputer, chip, etc.) or a processor (processor) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, ROM (Read-Only Memory), RAM (Random Access Memory), disk or optical disk and other media that can store program codes.

[0102] Those skilled in the art will appreciate that the above embodiments are specific embodiments for implementing the present application, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present application.

Claims

1. A one-dimensional range profile recognition method based on bidirectional temporal convolution, characterized in that: include: The obtained sample one-dimensional range image data is preprocessed, and after marking the real label, it is divided into training set and test set according to the preset ratio; A one-dimensional range image recognition model is constructed. The one-dimensional range image recognition model consists of a bidirectional temporal convolutional network as a feature extractor and a classifier. The bidirectional temporal convolutional network is used to perform forward temporal processing and backward temporal processing on the input one-dimensional range image data, establish a bidirectional temporal relationship between different distance units of the one-dimensional range image, and extract translation-invariant features. The classifier is used to perform category recognition based on the translation-invariant features and output recognition results. Based on the training set, supervised iterative training is performed on the one-dimensional range image recognition model to continuously optimize the network parameters of the one-dimensional range image recognition model until the one-dimensional range image recognition model converges and the optimal one-dimensional range image recognition model is obtained; The one-dimensional range image data of the mobile target to be identified is input into the optimal one-dimensional range image recognition model to obtain the recognition result of the one-dimensional range image data of the mobile target output by the optimal one-dimensional range image recognition model.

2. The one-dimensional range profile recognition method based on bidirectional temporal convolution according to claim 1 is characterized in that: The acquired sample one-dimensional range image data is preprocessed, including: De-noising the acquired sample one-dimensional range image data, smoothing the noise in the sample one-dimensional range image data to improve the separability between different categories; The denoised sample one-dimensional range image data is cut off with a preset length based on the center point of the target. If the length is insufficient, zero padding is performed to fill the preset length to obtain the preprocessed sample one-dimensional range image data.

3. The one-dimensional range profile recognition method based on bidirectional temporal convolution according to claim 1 is characterized in that: The bidirectional temporal convolutional network adopts the Bi-LSTM network structure, which is specifically composed of a forward LSTM layer, a backward LSTM layer and an output layer. The forward LSTM layer and the backward LSTM layer both contain multiple LSTM feature learning units. The forward LSTM layer and the backward LSTM layer obtain the temporal information in the input one-dimensional range image from two different time directions respectively. The forward LSTM layer is used to capture the context information from the starting point to the current distance unit and obtain the forward hidden state; The backward LSTM layer is used to capture the context information from the current distance unit to the end point and obtain the backward hidden state; The output layer is used to concatenate the forward hidden state and the backward hidden state to finally obtain the translation invariant features.

4. The one-dimensional range profile recognition method based on bidirectional temporal convolution according to claim 3 is characterized in that: The translation invariant features finally extracted by the bidirectional temporal convolutional network are expressed as follows: H n =[h1,h2,…,h n ,…,h N ]; Among them, H n represents the translation invariant features finally extracted by the bidirectional temporal convolutional network, Cat(·) represents the splicing of the channel dimension, and are the forward hidden state and backward hidden state of the nth distance unit, LSTM forward (·) represents the forward LSTM layer, LSTM backward (·) represents the backward LSTM layer, N represents the total number of range units, and prof(n) represents the echo strength of the nth range unit.

5. The one-dimensional range profile recognition method based on bidirectional temporal convolution according to claim 4 is characterized in that: The LSTM feature learning unit consists of an input gate, a forget gate, an output gate, and a memory unit. The input and output functions of the input gate, the forget gate, and the output gate are expressed by the formula: I n =σ[W hi h n-1 +W xi prof(n)+b i ]; f n =σ[W hf h n-1 +W xf prof(n)+b f ]; y n =σ[W hy h n-1 +W xy prof(n)+b y ]; Among them, prof(n) is the input of the input gate, forget gate and output gate, i n represents the output of the input gate, f n represents the output of the forget gate, y n represents the output of the output gate, σ represents the Sigmoid activation function, W xi , W xf and W xy is the weight matrix of the input gate, forget gate and output gate for the current input, W hi , W hf and W hy is the weight matrix of the input gate, forget gate and output gate for the previous input, b i 、b f and b y Bias for input gate, forget gate and output gate; Suppose the value stored in the memory cell at the current moment is C n , C n The calculation formula is: C n =C n-1 ×f n +tanh(W c [h n-1 ]+b c )×prof(n); Among them, C n-1 represents the value stored in the memory cell at the previous moment, tanh(n) represents the Tanh activation function, W c is the weight matrix of the memory unit for the current input, b c is the bias of the memory unit; The output of the LSTM feature learning unit at the current moment is expressed by the formula: in, Represents the output of the LSTM feature learning unit at the current moment.

6. The one-dimensional range profile recognition method based on bidirectional temporal convolution according to claim 1 is characterized in that: The classifier consists of a first fully connected layer, a ReLU activation layer, a second fully connected layer, and a Softmax layer connected sequentially. The classifier performs category recognition based on translation-invariant features and outputs the recognition result, which is expressed by the formula: f classifer (H n )=f softmax {f Linear2 [f ReLU (f Linear1 [H n ])]}; Among them, f classifer (·) represents the classifier, f Linear1 (·) represents the first fully connected layer, f ReLU (·) represents the ReLU activation layer, f Linear2 (·) represents the second fully connected layer, f softmax represents the Softmax layer, Represents the output recognition result.

7. The one-dimensional range profile recognition method based on bidirectional temporal convolution according to any one of claims 1 to 6, characterized in that: Based on the training set, supervised iterative training is performed on the one-dimensional range image recognition model to continuously optimize the network parameters of the one-dimensional range image recognition model until the one-dimensional range image recognition model converges to obtain the optimal one-dimensional range image recognition model, including: Inputting the sample one-dimensional range image data in the training set into the one-dimensional range image recognition model in batches for supervised training, and obtaining the recognition result of the sample one-dimensional range image data output by the one-dimensional range image recognition model; Using the cross entropy loss function, the loss value between the true label of the sample one-dimensional range image data and the recognition result of the sample one-dimensional range image data output by the one-dimensional range image recognition model is calculated; Based on the loss value, the stochastic gradient descent method is used to continuously optimize and adjust the network parameters of the one-dimensional range profile recognition model until the one-dimensional range profile recognition model converges to obtain the optimal one-dimensional range profile recognition model.

8. The one-dimensional range profile recognition method based on bidirectional temporal convolution according to claim 7 is characterized in that: The cross entropy loss function is used to calculate the loss value between the true label of the sample one-dimensional range image data and the recognition result of the sample one-dimensional range image data output by the one-dimensional range image recognition model, which is expressed by the formula: Where B represents the batch size, D represents the total number of categories, and CE(·) represents the cross entropy loss function. represents the true label of the one-dimensional distance image data of the b-th sample, It represents the recognition result of the one-dimensional range image data of the b-th sample output by the one-dimensional range image recognition model, and LOSS represents the calculated loss value.

9. A one-dimensional range profile recognition system based on bidirectional temporal convolution, characterized in that: include: The sample partitioning module is used to pre-process the acquired sample one-dimensional range image data, and after marking the real label, divide it into a training set and a test set according to a preset ratio; A model building module is used to build a one-dimensional range image recognition model, which is composed of a bidirectional temporal convolutional network as a feature extractor and a classifier; wherein the bidirectional temporal convolutional network is used to perform forward temporal processing and backward temporal processing on the input one-dimensional range image data, establish a bidirectional temporal relationship between different distance units of the one-dimensional range image, and extract translation-invariant features; the classifier is used to perform category recognition based on the translation-invariant features and output recognition results; A model training module is used to perform supervised iterative training on the one-dimensional range image recognition model based on the training set to continuously optimize the network parameters of the one-dimensional range image recognition model until the one-dimensional range image recognition model converges to obtain the optimal one-dimensional range image recognition model; The model using module is used to input the one-dimensional range image data of the mobile target to be identified into the optimal one-dimensional range image recognition model to obtain the recognition result of the one-dimensional range image data of the mobile target output by the optimal one-dimensional range image recognition model.

Images