A Ship Radar Recognition Method Based on Deep Learning

Abstract

The present invention provides a ship radar recognition method based on deep learning, which relates to the technical field of ship radar signal processing. The content includes: performing denoising and filtering processing on the echo signal reflected by the target, and generating spatial features and filtered target feature signals based on the denoised and filtered signal; performing enhancement and normalization processing on the filtered target feature signals, further performing spatio-temporal decomposition and non-linear fusion to obtain a fused spatio-temporal signal matrix; performing windowing processing on the fused spatio-temporal signal matrix and optimizing spatio-temporal features; generating data after spatio-temporal feature fusion through a spatio-temporal feature adaptive optimization algorithm; inputting the data after spatio-temporal feature fusion into a target recognition model, outputting the prediction result and confidence value of the target category, and optimizing the target recognition model. It solves the problems that traditional radar target recognition methods are difficult to effectively extract accurate spatio-temporal features, lack adaptability, and the generalization ability of the target recognition model is insufficient.

Description

Technical Field

[0001] The present invention relates to the technical field of ship radar signal processing, and particularly relates to a ship radar recognition method based on deep learning. Background Art

[0002] Radar target recognition technology has important applications in fields such as ship navigation and collision avoidance warning. However, its target recognition accuracy and efficiency still face challenges under complex sea conditions. Traditional radar target recognition methods rely on manually designed feature extraction and classification models, and it is difficult to adapt to the dynamics of target features changing over time and environment. The variability of signal noise, target motion, and complex sea conditions will significantly reduce the feature extraction ability of traditional radar target recognition methods, resulting in a decline in classification accuracy. In addition, due to the complex spatio-temporal characteristics of target signals, existing models often have difficulty effectively capturing the significant features in the signals, especially in scenarios with strong noise interference and drastic environmental changes.

[0003] In recent years, deep learning technology has provided new solutions for radar target recognition. Through multi-layer non-linear mapping, it can automatically extract the key features in the signals. However, deep learning models still have problems such as insufficient generalization ability, overfitting, and poor adaptability to changes in feature distributions in practical applications. At the same time, under changing environments, the spatio-temporal stability of target features is insufficient, and traditional deep learning models lack adaptive capabilities, making it difficult to handle the complex interactions between target signals and background noise. In addition, the sub-optimization of feature fusion methods and insufficient samples also limit the performance improvement of deep learning models. Therefore, there is an urgent need for an intelligent method that can adapt to complex environmental changes and efficiently extract target spatio-temporal features to improve the accuracy and efficiency of radar target recognition. Summary of the Invention

[0004] The present invention provides a ship radar recognition method based on deep learning to solve the problems in radar target recognition that due to signal noise and environmental changes, traditional target recognition methods are difficult to effectively extract accurate spatio-temporal features; under changing environments, the spatio-temporal features of targets are unstable, resulting in low classification accuracy; existing target recognition models lack adaptive capabilities and are difficult to cope with the dynamic changes of target features and environmental conditions; and the generalization ability of target recognition models is insufficient, prone to overfitting, resulting in unsatisfactory recognition effects in practical applications.

[0005] A ship radar recognition method based on deep learning includes the following steps:

[0006] S1. Perform denoising and filtering processing on the echo signal reflected by the target. Based on the denoised and filtered signal, generate spatial features and the filtered target feature signal; perform enhancement and normalization processing on the filtered target feature signal to obtain the normalized target feature signal;

[0007] S2. Perform spatio-temporal decomposition on the normalized target feature signal to obtain a spatio-temporal matrix, and through non-linear fusion, obtain a fused spatio-temporal signal matrix; perform windowing on the fused spatio-temporal signal matrix to obtain a windowed signal, and optimize the spatio-temporal features of the windowed signal, and finally construct a training dataset;

[0008] S3. Generate data after spatio-temporal feature fusion through a spatio-temporal feature adaptive optimization algorithm for target classification;

[0009] S4. Input the data after spatio-temporal feature fusion into the target recognition model, output the prediction result and confidence value of the target category, and optimize the target recognition model.

[0010] Preferably, the S2 specifically includes:

[0011] Decompose the normalized target feature signal into a spatio-temporal matrix through an improved spatio-temporal fluctuation decomposition and fusion algorithm, and the formula is as follows:

[0012] ,

[0013] where, is the spatio-temporal matrix, which is used to decompose the frequency features of the normalized target feature signal at time and azimuth angle ; is the normalized target feature signal at time ; is the wavelet basis matrix; is the translation parameter of the wavelet; is the scale parameter of the wavelet; is the frequency variable of the normalized target feature signal; is the direction weight; is the time variable; is the imaginary unit.

[0014] Preferably, the S2 specifically includes:

[0015] In the implementation process of windowing the fused spatio-temporal signal matrix, an adaptive dynamic windowing algorithm is adopted to define an adaptive windowing function, and the specific formula is:

[0016] ,

[0017] where, is the adaptive windowing function; is the square of the gradient of the fused spatio-temporal signal matrix; is the discrete time index; is the azimuth angle; is the window adjustment parameter; is the spatio-temporal signal matrix after fusion.

[0018] Preferably, the S2 specifically includes:

[0019] In the process of optimizing the spatio-temporal features of the windowed signal, extract and standardize the key spatio-temporal features of the windowed signal to obtain the standardized key spatio-temporal features, and expand the standardized key spatio-temporal features into high-dimensional feature vectors; construct a training dataset based on the high-dimensional feature vectors and the labels of the targets.

[0020] Preferably, the S3 specifically includes:

[0021] In the implementation process of the spatio-temporal feature adaptive optimization algorithm, fuse the spatio-temporal features of the target in a weighted manner. The specific formula is as follows:

[0022] ,

[0023] where, represents the data after spatio-temporal feature fusion; represents the function of the spatio-temporal feature fusion operation; is the high-dimensional feature vector; represents the th time feature of the target; represents the th spatial feature of the target; and are the adaptive weights of the time and spatial features respectively; represents the element-wise dot product operation; represents the product operation on the spatial features; is the total number of time features; is the total number of spatial features.

[0024] Preferably, the S3 specifically includes:

[0025] Introduce an adaptive weighting mechanism to optimize the adaptive weights of the time and spatial features.

[0026] Preferably, the S3 specifically includes:

[0027] Based on the data after spatio-temporal feature fusion, use a multi-layer perceptron for classification to obtain the probabilities that the target belongs to different classes; at the same time, introduce a regularization term to constrain the parameters of the multi-layer perceptron; based on the regularization term and the cross-entropy loss function, generate a total loss function, and train the multi-layer perceptron by minimizing the total loss function.

[0028] Preferably, the S4 specifically includes:

[0029] Optimize the target recognition model through a feedback mechanism, which includes two parts: one is error correction based on the current recognition result, and the other is dynamic adjustment of the target recognition model based on environmental changes.

[0030] The beneficial effects of the technical solution of the present invention are as follows:

[0031] 1. Through multi-stage signal processing, including denoising, wavelet transform, empirical mode decomposition, frequency domain analysis, and time domain feature extraction, effectively extract the spatio-temporal features of the target. Eliminate the dimensionality influence between different features through normalization processing to ensure the unity and reliability of the features, and improve the efficiency and accuracy of subsequent target recognition and analysis.

[0032] 2. Through the improved spatio-temporal fluctuation decomposition and fusion algorithm and the adaptive dynamic windowing algorithm, enhance the dynamics and robustness of target recognition. The adaptive dynamic windowing algorithm adjusts the window size according to the change of the target feature signal to improve the clarity and recognizability of the target feature signal. Expand the training samples through data augmentation to enhance the generalization ability of the deep learning model, and finally improve the anti-interference ability, accuracy, and target recognition efficiency of the deep learning model.

[0033] 3. Combine the spatio-temporal feature adaptive optimization algorithm with the deep learning model to achieve efficient recognition and classification of radar targets. By weighted fusion of spatio-temporal features, enhance the information interaction and feature expressiveness, and improve the classification accuracy. Input the data after spatio-temporal feature fusion into the deep learning model to achieve target prediction and confidence generation, and optimize the target recognition model through incremental learning and feedback mechanism. Dynamically adjust the recognition window size through the dynamic windowing mechanism to ensure the accuracy of feature extraction, and improve the adaptability and recognition accuracy of the system in different environments. Brief Description of the Drawings

[0034] Figure 1 It is a flowchart of a ship radar recognition method based on deep learning according to the present invention. Detailed Embodiments

[0035] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0037] The following specifically describes the specific solution of a ship radar recognition method provided by the present invention in conjunction with the accompanying drawings.

[0038] Refer to the attached Figure 1 , which shows a ship radar recognition method based on deep learning provided by an embodiment of the present invention. The method includes the following steps:

[0039] S1. Denoise and filter the echo signal reflected by the target. Based on the denoised and filtered signal, generate spatial features and the filtered target feature signal; enhance and normalize the filtered target feature signal to obtain the normalized target feature signal.

[0040] Receive the echo signal reflected by the target through the radar antenna , and this signal is a continuous-time signal. Use an analog-to-digital converter (ADC) to sample it, and the sampling period is . According to the sampling result, obtain the discretized time signal .

[0041] ,

[0042] Among them, represents analog-to-digital conversion; is the echo signal reflected by the target; is the sampling period, which determines the time resolution after discretization; is the discrete time index, indicating the position of the sampling point; is the discretized radar signal; is the sampling pulse signal, used to extract the signal value at time .

[0043] Use wavelet transform to decompose into different scales, obtain its high-frequency component and low-frequency component, remove the main noise part in the high-frequency component, and retain the low-frequency component. Use inverse wavelet transform to reconstruct the processed components into the denoised signal .

[0044] ,

[0045] Among them, is the wavelet transform operator; is the radar signal discretized at time ; is the mother wavelet function; is the scale factor, indicating the frequency range of decomposition; is the translation factor, indicating the movement in time.

[0046] For the denoised signal Perform step-by-step decomposition through empirical mode decomposition (EMD) to obtain several intrinsic mode functions (IMFs) and a residual signal. Remove the influence of high-frequency noise and reconstruct the low-frequency effective components to obtain the denoised and filtered signal .

[0047] ,

[0048] wherein, is the th intrinsic mode function, which is used to reflect different frequency components of the signal; is the residual signal, representing the undecayed part; is the total number of intrinsic mode functions obtained by decomposition.

[0049] Apply the fast Fourier transform to the denoised and filtered signal to convert it from the time domain to the frequency domain and extract spectral features .

[0050] ,

[0051] wherein, is the frequency; is the input time-domain signal; is the imaginary unit.

[0052] Apply windowing operations to extract the time-domain characteristics of the denoised and filtered signal , such as peak value, average value, etc., calculate the characteristics of the target in time, and extract time-domain features .

[0053] ,

[0054] wherein, is the denoised and filtered signal; is the windowing function, which is used to extract local time characteristics.

[0055] Map the denoised and filtered signal to the range-azimuth diagram to extract the spatial position information of the target and generate spatial features .

[0056] ,

[0057] wherein, is the target azimuth angle; is the target distance; is to map to the two-dimensional coordinate system to generate a distribution diagram for visualizing the spatial position information of the target.

[0058] Design a set of band-pass filters to retain the frequency range of the target feature signal. The target feature signal refers to the feature part related to the target in the echo signal reflected by the target received by the radar, specifically the feature frequency components related to the target in the echo signal reflected by the target received by the radar, including the effects of target motion and characteristics (such as speed, distance, azimuth angle, scattering characteristics, etc.) on the spectrum of the echo signal reflected by the target received by the radar. The filter design uses the window function method or the frequency domain sampling method to remove the noise of the non-target feature signal frequency and highlight the frequency components of the target feature signal to ensure that the characteristics of the target feature signal are not affected by excessive smoothing.

[0059] ,

[0060] Among them, is the filtered target feature signal; are the coefficients of the filter, which determine the frequency characteristics of the filter; is the signal after denoising and filtering at time ; is the delay factor, indicating the response length of the filter.

[0061] Then, use histogram equalization or adaptive contrast enhancement method to enhance the difference between the target and the background in the target feature signal. By stretching the dynamic range of the filtered target feature signal, the target recognition ability in low-contrast environments is enhanced.

[0062] ,

[0063] Among them, is the enhanced target feature signal; , are the minimum and maximum values of the filtered target feature signal; , is the target dynamic range of the filtered target feature signal.

[0064] Perform linear normalization on the enhanced target feature signal to adjust its value to the standard range and eliminate the dimensional influence between different features.

[0065] ,

[0066] Among them, is the normalized target feature signal; , are the minimum and maximum values of the enhanced target feature signal.

[0067] S2. Perform spatio-temporal decomposition on the normalized target feature signal to obtain a spatio-temporal matrix, and through non-linear fusion, obtain a fused spatio-temporal signal matrix; perform windowing on the fused spatio-temporal signal matrix to obtain a windowed signal, and optimize the spatio-temporal features of the windowed signal, and finally construct a training data set.

[0068] Introduce the spatio-temporal fluctuation decomposition and fusion algorithm and the adaptive dynamic windowing algorithm, extract the dynamic spatio-temporal features of the normalized signal, and combine data augmentation and normalization processing to construct a multi-dimensional feature data set;

[0069] Perform spatio-temporal decomposition on the normalized target feature signal through an improved spatio-temporal fluctuation decomposition and fusion algorithm to separate the time-frequency features in the target feature signal and perform multi-scale fusion.

[0070] The normalized target feature signal is decomposed into a spatio-temporal matrix to capture the details of the normalized target feature signal changing with time and space.

[0071] ,

[0072] where is the spatio-temporal matrix, used to decompose the frequency features of the normalized target feature signal at time and azimuth ; is the normalized target feature signal at time ; is the wavelet basis matrix, representing the kernel function of the multi-scale wavelet transform, originating from the wavelet transform theory; is the translation parameter of the wavelet; is the scale parameter of the wavelet; is the frequency variable of the normalized target feature signal, related to the time-varying characteristics of the signal; is the direction weight, used to describe the characteristics of the normalized target feature signal in different directions; is the time variable, representing the sampling time point; is the imaginary unit, appearing in the complex exponential, used to represent the frequency characteristics of the normalized target feature signal; is the frequency component of the normalized target feature signal, originating from the Fourier transform.

[0073] Through non-linear fusion, enhance the significant features in the spatio-temporal matrix while suppressing the interference of background noise.

[0074] ,

[0075] where is the spatio-temporal signal matrix after fusion; is the number of wavelet scales, indicating different scales or frequency decomposition levels of the wavelet functions used in the spatio-temporal decomposition process; is the number of wavelet translations, indicating the number of times or step sizes of translating the wavelet function in the wavelet transform; is the fusion weight, dynamically calculated from the frequency response of the spatio-temporal matrix; ) is to take the logarithm of the square of the modulus of the spatio-temporal matrix to enhance the weak signal response; is used to smooth the amplitude variation of the signal; is the square of the two-norm of the spatio-temporal matrix.

[0076] Adopt an adaptive dynamic windowing algorithm to adjust the window size according to the change of the target feature signal after preprocessing, and further optimize the target feature extraction. Define an adaptive windowing function so that the window size is inversely proportional to the signal change intensity. A small window is used at the fast-changing part, and a large window is used at the slow-changing part.

[0077] ,

[0078] where, is the adaptive windowing function, used to obtain the signal after adaptive windowing; is the square of the gradient of the spatio-temporal signal matrix after fusion, used to describe the change rate of the target; is the discrete time index; is the azimuth angle; is the window adjustment parameter, related to the change characteristics of the target; is used to enhance the significance of the spatio-temporal signal matrix after fusion; is the normalization factor, used to ensure the consistency of the adaptive windowing function.

[0079] Highlight the dynamic characteristics of the signal after adaptive windowing through dynamic windowing and eliminate the interference of low-frequency background components. The specific implementation formula of dynamic windowing is:

[0080] ,

[0081] where, is the signal after windowing.

[0082] Next, optimize the spatio-temporal features of the windowed signal. Specifically, extract and standardize the key spatio-temporal features of the windowed signal for constructing the input of the deep learning model. Specifically, normalize the key spatio-temporal features through non-linear mapping to enhance the sparsity and robustness of the windowed signal, so that the deep learning model has strong anti-interference ability against noise.

[0083] ,

[0084] Among them, is the key spatio-temporal feature after standardization, which is used to improve the signal sparsity and robustness and enhance the anti-interference ability of the deep learning model.

[0085] Expand the standardized key spatio-temporal features into high-dimensional feature vectors, and the feature vectors contain the comprehensive characteristics of the target at different azimuth angles and time scales.

[0086] ,

[0087] Among them, is the high-dimensional feature vector, representing the spatio-temporal features of the target; is the total dimension of the target, containing spatio-temporal dynamic information; is the target dimension index, ; is used to expand the standardized key spatio-temporal features to different azimuth angles.

[0088] Then, data augmentation is performed by means of rotation, scaling, noise, etc. to expand the training samples. After data augmentation, normalization is performed again to ensure the consistency of spatio-temporal features and target feature signals.

[0089] ,

[0090] Among them, is the final signal after enhancement, containing noise perturbation and rotation; , are random noise parameters, which are used to simulate the real environment; is used to simulate the azimuth change of the signal; is used to simulate the time periodic change of the signal.

[0091] Finally, construct the training data set :

[0092] ,

[0093] Among them, is the label of the target, containing information such as target type, trajectory, speed, etc.

[0094] S3. Through the spatio-temporal feature adaptive optimization algorithm, generate the data after spatio-temporal feature fusion and perform target classification.

[0095] In order to better classify the target, design a spatio-temporal feature adaptive optimization algorithm, combine the time and space features, and enhance the accuracy of target classification.

[0096] Through the fusion of spatio-temporal features, the temporal information and spatial information of the target are effectively interacted. Assume the spatio-temporal features of the target contain temporal features and spatial features, and these features are fused in a weighted manner.

[0097] ,

[0098] wherein, represents the data after spatio-temporal feature fusion; represents the function of spatio-temporal feature fusion operation; represents the th temporal feature of the target, usually including speed, acceleration, etc.; represents the th spatial feature of the target, usually including scattering intensity, azimuth angle, elevation angle, etc.; and are the adaptive weights of temporal and spatial features respectively, obtained through training; represents the element-wise dot product operation; represents the product operation on spatial features, adding more complex non-linear transformations; is the normalization process of spatial features to avoid numerical instability.

[0099] To further optimize the fusion effect of spatio-temporal features, an adaptive weighting mechanism is introduced, and by optimizing the weights, it learns how to weight the contribution of each feature.

[0100] ,

[0101] ,

[0102] wherein, represents the th temporal feature of the target; represents the th temporal feature of the target; represents the L2 norm of the spatio-temporal features of the target; the adaptive weight formula of temporal features uses an exponential function to non-linearly optimize the weighting of temporal features; the adaptive weight formula of spatial features uses the Euclidean distance to calculate the similarity between spatial features as the basis for weighting.

[0103] The data after spatio-temporal feature fusion is used for target classification using deep learning methods. Assume a multi-layer perceptron (MLP) is used to complete the classification task, and the output of each layer undergoes a non-linear transformation through an activation function.

[0104] Assume the input of the multi-layer perceptron is , and the output of each layer can be expressed as .

[0105] Among them, is the output of the -th layer; is the output of the -th layer; and are the weights and biases of the -th layer respectively; is the activation function, and common activation functions include ReLU, Sigmoid or Tanh.

[0106] At the last layer of the multi-layer perceptron, the probabilities that the target belongs to different classes are obtained.

[0107] ,

[0108] Among them, is the predicted probability of the target class after fusing the given spatio-temporal features of the data , and are the weights and biases of the output layer; is the number of classes; is the output of the -th layer; represents the output of the -th layer, the -th class; is the index of the class; refers to the last layer of the multi-layer perceptron.

[0109] To train the target classification model, that is, the multi-layer perceptron, the cross-entropy loss function is used to measure the difference between the output of the target classification model and the true label.

[0110] ,

[0111] Among them, is the cross-entropy loss function, which is used to measure the difference between the model output and the true label; is the true label of the target; represents that after the given input features , the model predicts the probability that the -th sample belongs to the target class ; is the number of samples, that is, the total number of samples in the training dataset.

[0112] At the same time, to avoid overfitting, a regularization term is introduced, and L2 regularization is used to constrain the parameters of the target classification model:

[0113] ,

[0114] Among them, is the L2 regularization term, which is used to prevent overfitting; is the regularization coefficient, which is used to control the influence of the regularization term; is the L2 norm of the weight matrix, which is used to penalize large weights.

[0115] The final total loss function is:

[0116] ,

[0117] By minimizing the total loss function , the target classification model can be trained.

[0118] Use the gradient descent method to optimize the target classification model and update the weights and biases. Calculate the gradients through the chain rule, and the parameter update formula is:

[0119] ,

[0120] ,

[0121] Among them, is the learning rate; and are the partial derivatives of the loss function with respect to and respectively.

[0122] S4. Input the data after spatio-temporal feature fusion into the target recognition model, output the prediction results of the target category and the confidence values, and optimize the target recognition model.

[0123] Input the data after spatio-temporal feature fusion obtained from S3 into a deep learning model. Through a data stream processing system, such as Apache Kafka or Apache Flink, receive real-time data from devices such as sensors and cameras, and the real-time data will be input into the target recognition model in real time, such as a multi-layer perceptron, to generate the target category and its confidence at the current moment.

[0124] The model outputs the prediction results of the target category and its corresponding confidence values, such as classification probabilities.

[0125] ,

[0126] Among them, is the final input feature after spatio-temporal feature optimization, Is the inference function of the deep learning model.

[0127] The identified targets and confidence levels are used to further optimize the target recognition model through a feedback mechanism. The feedback mechanism consists of two parts: one is error correction based on the current recognition results, and the other is dynamic adjustment of the target recognition model based on environmental changes.

[0128] The error between the real-time monitored target recognition results and the actual situation can be monitored through sensors for the actual state of the target or confirmed through manual feedback for recognition accuracy. If the system detects an error (e.g., misclassification, low confidence, etc.), the feedback mechanism will be immediately triggered.

[0129] If a recognition error is found, an incremental learning method is used to dynamically adjust the target recognition model. The newly obtained data is used to fine-tune the target recognition model to reduce future errors.

[0130] The formula for the incremental learning method is:

[0131] ,

[0132] where, are the adjusted model parameters; are the model parameters; is the learning rate; is the gradient of the loss function; are the input features, is the target label; represents the loss function, which is determined according to the specific implementation scenario.

[0133] Based on the feedback error, the target recognition model is updated in an adaptive manner. For example, the hyperparameters such as the learning rate, number of layers, regularization, etc. of the target recognition model can be adjusted according to the feedback, so that the target recognition model can better handle the changes in target features. While optimizing the feedback, the performance of the target recognition model is evaluated in real time to ensure the computational efficiency and accuracy in real-time scenarios. Evaluation metrics such as precision, recall, and F1-score can be used to dynamically monitor the performance of target recognition.

[0134] To adapt to different environmental changes and target features, the dynamic windowing mechanism can adjust the size of the recognition window according to the actual situation. This process dynamically adjusts the size of the input data window based on the movement speed, scale change, and environmental noise of the target. By monitoring information such as the movement speed and size change of the target, the size of the input window is dynamically adjusted to ensure the accuracy of feature extraction.

[0135] Window size update:

[0136] ,

[0137] Among them, is the current window size, is the window size at the previous moment, is the window adjustment amount.

[0138] If the environment changes significantly (such as changes in light, background noise, etc.), the threshold can be set through the expert experience method to automatically trigger the adjustment of the window.

[0139] Change detection:

[0140] ,

[0141] Among them, is the current environmental state (such as light, noise, etc.); is the environmental state at the previous moment; is the change amount.

[0142] According to the movement speed and size of the target, convolution kernels of different scales are automatically selected for processing or appropriate-sized windows are used in the LSTM to capture the dynamic features in time, so as to ensure that the target can be effectively recognized at different scales.

[0143] In summary, a ship radar recognition method based on deep learning is completed.

[0144] The sequence of the invention embodiments is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be beneficial.

[0145] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and the key points of each embodiment are the differences from other embodiments.

[0146] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; 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: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on 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 each embodiment of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A ship radar identification method based on deep learning, characterized in that: The following steps are involved: S1. De-noising and filtering the echo signal reflected by the target, and generating spatial features and filtered target feature signals based on the de-noising and filtered signals; Performing enhancement and normalization processing on the filtered target feature signal to obtain a normalized target feature signal; S2. Performing spatiotemporal decomposition on the normalized target feature signal to obtain a spatiotemporal matrix, and obtaining a fused spatiotemporal signal matrix through nonlinear fusion; performing windowing processing on the fused spatiotemporal signal matrix to obtain a windowed signal, and optimizing the spatiotemporal features of the windowed signal, and finally constructing a training data set; S3. Generate data after fusion of spatiotemporal features through spatiotemporal feature adaptive optimization algorithm to perform target classification; S4. Input the data after fusion of spatiotemporal features into the target recognition model, output the prediction results and confidence values ​​of the target category, and optimize the target recognition model.

2. The ship radar identification method based on deep learning according to claim 1 is characterized in that: The S2 specifically includes: Through the improved space-time fluctuation decomposition and fusion algorithm, the normalized target feature signal is decomposed into a space-time matrix, and the formula is as follows: ， in, is a spatiotemporal matrix used to decompose the normalized target feature signal at time and azimuth Frequency characteristics on ; It is at the moment Normalized target feature signal; is the wavelet basis matrix; is the translation parameter of the wavelet; is the scale parameter of the wavelet; is the frequency variable of the normalized target feature signal; is the direction weight; is a time variable; Is an imaginary unit.

3. The ship radar identification method based on deep learning according to claim 1, characterized in that: The S2 specifically includes: In the process of implementing windowing processing on the fused space-time signal matrix, an adaptive dynamic windowing algorithm is used to define an adaptive windowing function. The specific formula is: ， in, is the adaptive windowing function; is the square gradient of the fused spatiotemporal signal matrix; is a discrete time index; is the azimuth; Adjust parameters for the window; is the fused spatiotemporal signal matrix.

4. The ship radar identification method based on deep learning according to claim 1, characterized in that: The S2 specifically includes: In the process of optimizing the spatiotemporal features of the windowed signal, the key spatiotemporal features of the windowed signal are extracted and standardized to obtain the standardized key spatiotemporal features, and the standardized key spatiotemporal features are expanded into a high-dimensional feature vector; based on the high-dimensional feature vector and the target label, a training data set is constructed.

5. The ship radar identification method based on deep learning according to claim 1, characterized in that: The S3 specifically includes: In the process of implementing the spatiotemporal feature adaptive optimization algorithm, the spatiotemporal features of the target are fused in a weighted manner. The specific formula is as follows: ， in, Represents the data after the fusion of spatiotemporal features; A function representing the spatiotemporal feature fusion operation; is a high-dimensional feature vector; The target time characteristics; The target spatial features; and are the adaptive weights of temporal and spatial features, respectively; Represents an element-by-element dot product operation; Represents the product operation on spatial features; is the total number of time features; is the total number of spatial features.

6. The ship radar identification method based on deep learning according to claim 5 is characterized in that: The S3 specifically includes: An adaptive weighting mechanism is introduced to optimize the adaptive weights of temporal and spatial features.

7. The ship radar identification method based on deep learning according to claim 5, characterized in that: The S3 specifically includes: Based on the data after the fusion of spatiotemporal features, a multi-layer perceptron is used for classification to obtain the probability that the target belongs to different categories; at the same time, a regularization term is introduced to constrain the parameters of the multi-layer perceptron; based on the regularization term and the cross entropy loss function, a total loss function is generated, and the multi-layer perceptron is trained by minimizing the total loss function.

8. The ship radar identification method based on deep learning according to claim 1, characterized in that: The S4 specifically includes: The target recognition model is optimized through a feedback mechanism, which consists of two parts: one is error correction based on the current recognition results, and the other is dynamic adjustment of the target recognition model based on environmental changes.

Images