RecycleNet-based sea surface target detection and classification method and system

Abstract

The invention discloses a RecycNet-based sea surface target detection and classification method and system, and belongs to the technical field of radar target classification. The RecycNet-based sea surface target detection and classification method comprises the following steps: acquiring a sea surface echo amplitude sequence by using a radar, and processing the sea surface echo amplitude sequence by using a recurrence plot construction technology to obtain recurrence plot data; the method comprises the following steps: constructing a RecycNet network; training a RecycNet network by using the recurrence plot data, and adjusting network parameters; performing classification by using the trained RecycNet network to obtain a classification result; compared with the traditional convolutional neural network, the RecycNet can more efficiently capture and recover the detail characteristics of the input data, so that the network shows stronger detection and classification capabilities under the complex background of sea clutters, and the detection precision and robustness are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of radar target classification, and in particular relates to a sea surface target detection and classification method and system based on RecycleNet. Background Art

[0002] Target detection in the sea clutter background is of great significance in both military and civilian fields, and the detection of small targets on the sea surface is one of the key research directions. However, due to the interference of sea clutter caused by the complex marine environment and the diversity of sea surface target types, achieving reliable and robust sea surface target detection has always been a key technical problem that needs to be further studied.

[0003] In response to the problem of target detection in the background of sea clutter, scholars have done a lot of research work, proposed many detection methods and achieved certain results. In 2022, Su Ningyuan et al. proposed a maritime target detection method based on radar signal graph data and graph convolution (GCN) in their paper "Maritime Target Detection Based on Radar Graph Data and Graph Convolutional Network". In 2019, Ai Jiaqiu et al. proposed a ship detection algorithm based on multi-scale rotation-invariant Haar-like (MSRI-HL) feature integrated convolutional neural network (MSRIHL-CNN) in their paper "Multi-Scale Rotation-Invariant Haar-Like Feature Integrated CNN-Based Ship Detection Algorithm of Multiple-Target Environment in SAR Imagery". However, deep learning algorithms generally have the problem of information loss, and it is impossible to avoid the loss of some key information in the feature extraction process.

[0004] This problem is particularly prominent in current convolutional neural networks (CNNs). Although different CNN architectures (such as AlexNet, Conv64F or ResNet12, etc.) have achieved a certain degree of success in learning discriminative features from input images to handle classification problems, the following problem still exists: the current CNN filtering will inevitably discard some feature information, and whether this discarded feature information is conducive to improving classification performance is still an open question. Summary of the invention

[0005] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a sea surface target detection and classification method and system based on RecycleNet, which solves the problems in the prior art.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A sea surface target detection and classification method based on RecycleNet includes the following steps:

[0008] The sea surface echo amplitude sequence is obtained by radar, and the recursive graph construction technology is used to process the recursive graph data;

[0009] Build the RecycleNet network;

[0010] Using the recursive graph data to train the RecycleNet network and adjust network parameters;

[0011] Use the trained RecycleNet network for classification to obtain the classification results.

[0012] Furthermore, the sea surface echo amplitude sequence is composed of N continuous pulse echo data received by the radar, forming a time series data set, and each pulse echo data contains its corresponding amplitude information.

[0013] Furthermore, the RecycleNet network includes multiple Recycle modules and a fully connected module; the Recycle module includes a convolutional layer, a Leaky ReLU activation layer, a maximum pooling layer and a subtraction layer for feature recycling; multiple Recycle modules are sequentially connected layer by layer to capture detailed features in the recursive graph through feature extraction and recycling mechanism.

[0014] Furthermore, a fully connected layer is set after each Recycle module to integrate the recycling feature information θ of each Recycle module. k (X), k = 1, 2, ... K are sent to the fully connected layer to merge the information. The weight matrix of the fully connected layer is W k , the bias term is b k , then the output formula through the fully connected layer is:

[0015] y k =W k θ k (X)+b k ,k=1,2,…K

[0016] The feature θ after each recycling k(X) The output information after the fully connected layer is merged, and the merged output comprehensive feature map y can be expressed as:

[0017]

[0018] Where K is the total number of outputted recovery feature information.

[0019] Furthermore, the final module of the RecycleNet network is a fully connected module, which receives the comprehensive feature map y and maps it to the classification space; the output of the fully connected module is a score, for each category i, let the score be y i ; The output of the fully connected module is converted into probability values ​​of each category through the softmax activation function. Assuming that the total number of categories is C, the probability of the i-th category is:

[0020]

[0021] in, represents the score of the i-th category.

[0022] Furthermore, the steps of classification using the RecycleNet network include:

[0023] S41, preparing test data to be classified;

[0024] S42, inputting the test data into the trained RecycleNet network;

[0025] S43, applying a softmax activation function to the output of the RecycleNet network to generate classification probabilities for each category;

[0026] S44, according to the output probability of the softmax activation function, the category with the highest probability is determined as the classification result.

[0027] A sea surface target detection and classification system based on RecycleNet, comprising:

[0028] Radar signal receiving and processing module: Use radar to obtain the sea surface echo amplitude sequence, and use recursive graph construction technology to process and obtain recursive graph data;

[0029] Network building module: building the RecycleNet network;

[0030] Network training module: using the recursive graph data to train the RecycleNet network and adjust network parameters;

[0031] And, classification module: use the trained RecycleNet network to perform classification and obtain the classification results.

[0032] A computer storage medium stores a readable program, which can execute the above-mentioned sea surface target detection and classification method based on RecycleNet when the program is run.

[0033] An electronic device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus;

[0034] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above-mentioned sea surface target detection and classification method based on RecycleNet.

[0035] A computer program product includes computer instructions, wherein the computer instructions instruct a computing device to perform operations corresponding to the above-mentioned sea surface target detection and classification method based on RecycleNet.

[0036] Beneficial effects of the present invention:

[0037] The present invention proposes a sea surface target detection and classification method based on RecycleNet, which obtains the sea surface echo amplitude sequence through radar, converts the one-dimensional data into two-dimensional recursive graph data using recursive graph construction technology, and then inputs the recursive graph data into the RecycleNet network for training and classification. RecycleNet is composed of multiple Recycle modules and fully connected modules. The core advantage of the present invention lies in the information recycling mechanism of RecycleNet, which recursively restores the detailed feature information lost in the convolution and pooling process through the subtraction recovery layer in the Recycle module multiple times, thereby ensuring the integrity and depth expression of the features. Compared with traditional convolutional neural networks, RecycleNet can more efficiently capture and restore the detailed characteristics of the input data, so that the network can show stronger detection and classification capabilities under the complex background of sea clutter, and significantly improve the detection accuracy and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 It is a flow chart of the sea surface target detection and classification method of the present invention;

[0040] Figure 2is a framework diagram of the RecycleNet and CommonNet networks in Example 2;

[0041] Figure 3 It is a structural diagram of the sea surface target detection and classification system of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] Example 1

[0044] like Figure 1 As shown, a sea surface target detection and classification method based on RecycleNet includes the following steps:

[0045] S1, using radar to obtain the sea surface echo amplitude sequence, and using recursive graph construction technology to process and obtain recursive graph data;

[0046] The sea surface echo amplitude sequence is composed of N continuous pulse echo data received by the radar, forming a time series data set, and each pulse echo data contains its corresponding amplitude information. The sea surface echo amplitude sequence is processed using a recursive graph construction technology to obtain recursive graph data, and the specific steps include:

[0047] By using the existing recursive graph generation method, the sea surface echo sequence is converted into a recursive graph through data segmentation, phase space reconstruction and grayscale normalization, and its nonlinear dynamic characteristics are extracted for classification; this method can convert the sea surface echo amplitude sequence data into a recursive graph;

[0048] S2, build the RecycleNet network;

[0049] The RecycleNet network consists of K-1 Recycle modules and a fully connected module; the Recycle module includes a convolutional layer, a Leaky ReLU activation layer, a maximum pooling layer, and a subtraction layer for feature recycling; the Recycle module is connected layer by layer, and captures the detailed features in the recursive graph through feature extraction and recycling mechanisms, providing rich information for classification tasks.

[0050] The RecycleNet network includes the following features and contents:

[0051] 1) In the Recycle module, a convolutional layer is configured for preliminary feature extraction. The convolutional layer uses a combination of 1×3 and 3×1 asymmetric convolutional kernels to extract horizontal and vertical features of the recurrence graph respectively. For the input image X, Ω1=X, the convolutional layer generates a preliminary feature map Γ1 through operation f as follows:

[0052] Γ1=f(Ω1)

[0053] This operation captures the local edge and texture features of the recursive graph, providing a basic feature expression for subsequent feature recovery.

[0054] 2) The Leaky ReLU activation layer in the Recycle module is used to realize nonlinear mapping. The Leaky ReLU activation layer is configured after the convolution layer. The activation layer enhances the expression ability of the feature map through nonlinear transformation and retains the negative part of the information, which helps to improve the network's recognition effect on small targets.

[0055] 3) The maximum pooling layer in the Recycle module uses a 2×2 maximum pooling operation to select the maximum value from each 2×2 region, thereby reducing the size of the feature map and reducing the amount of calculation. The combined operation of activation and pooling is defined as g, and the operation function that completes a set of three steps of convolution, activation, and pooling is h. This operation further compresses the size of the feature map while retaining key information, creating conditions for subsequent feature recycling.

[0056] 4) In the Recycle module, the feature information lost in the convolution process is recovered through the subtraction layer. The input X of the convolution layer is subtracted from the output Γ1 to obtain the first recycled feature information θ1(X):

[0057] θ1(X)=Ω1-Γ1=Ω1-f(Ω1)

[0058] Among them, θ1(X) contains the detailed features lost during the convolution process, providing preliminary feature recovery for subsequent recursive recovery.

[0059] 5) The Recycle module implements multiple rounds of feature recycling through a recursive mechanism. The feature information of the kth recycling is:

[0060] θ k (X) = Ω k -Γ k =Ω k -f(Ω k ),k=1,2,…K-1

[0061] The characteristic information of the last recycling is:

[0062] θ K(X) = h(Ω K-1 )

[0063] Through multiple rounds of recursive recycling, the Recycle module gradually restores the multi-level information lost during the convolution and pooling process to ensure the integrity of the feature map.

[0064] 6) RecycleNet connects multiple Recycle modules in sequence to form a layer-by-layer progressive feature extraction and information recycling structure. Assuming that RecycleNet contains K-1 Recycle modules, the input data is processed by convolution, activation, pooling and feature recycling of each Recycle module to extract and restore feature information layer by layer to achieve deep expression of the input image.

[0065] 7) RecycleNet sets a fully connected layer after each Recycle module to collect the recycling feature information of each Recycle module k (X), k = 1, 2, ... K are sent to the fully connected layer to merge the information. The weight matrix of the fully connected layer is W k , the bias term is b k , then the output formula through the fully connected layer is:

[0066] y k =W k θ k (X)+b k ,k=1,2,…K

[0067] The feature θ after each recycling k (X) The output information after the fully connected layer is merged (concatenation layer), and the merged output comprehensive feature map y can be expressed as:

[0068]

[0069] The concatenation layer integrates feature information from different levels into a comprehensive representation, thereby improving the classification performance.

[0070] 8) The final module of RecycleNet is a fully connected module that receives the comprehensive feature map y of the concatenated layer and maps it to the classification space. The output of the fully connected module is a score. For each category i, let the score be y i ; The output of the fully connected module is converted into probability values ​​of each category through the softmax activation function. Assuming that the total number of categories is C, the probability of the i-th category is:

[0071]

[0072] in, Represents the score of the i-th category. Through the probability distribution of the softmax output, RecycleNet can determine the classification result, thereby achieving accurate detection and classification of small targets.

[0073] 9) A loss layer is set up in RecycleNet. The output results and labels are sent to the loss layer, the loss value is calculated, and it is fed back to the network for parameter update, thereby optimizing the performance of the entire network.

[0074] S3, using the recursive graph data to train the RecycleNet network, and adjusting network parameters to improve the classification accuracy of small objects;

[0075] The steps to train the RecycleNet network include:

[0076] S31, dividing the obtained recursive graph data set into two parts, a training set and a test set, wherein the ratio of the training set to the test set is 7:3;

[0077] S32, labeling the divided recursive graph data respectively, wherein the labels are divided into two categories: target and clutter; inputting the labeled training set and test set into the RecycleNet network for training;

[0078] S33, optimizing the parameters of the RecycleNet network based on the training data so that RecycleNet can accurately distinguish the target and clutter features; during the training process, the RecycleNet network gradually minimizes the error between the predicted output and the actual label by continuously adjusting the parameters;

[0079] The parameters in the training process are set as follows: the input image size is 224×224, the initial learning rate is 0.001, the optimization function is Adam, the Epoch is set to 40, and the batch_size is set to 20, so as to obtain good network performance under moderate training times and batch sizes. Each Recycle module contains a 1×3 and a 3×1 convolution layer, the number of convolution kernels is 8, and is connected to a ReLU activation layer, followed by a 2×2 maximum pooling layer with a step size of 2, and finally connected to a fully connected layer containing 32 units.

[0080] After training, the performance of the RecycleNet network is evaluated using the test set to verify the classification accuracy and generalization ability of the network in small target detection. The evaluation indicator is the F1 score value, and its formula is:

[0081]

[0082] Among them, the definitions of Precision and Recall are: And TP is the number of samples correctly predicted as the target category; FP is the number of samples incorrectly predicted as the target category; FN is the number of samples not correctly identified as the target category.

[0083] S4, use the trained RecycleNet network to perform classification and obtain the classification results.

[0084] The classification steps include:

[0085] S41, prepare test data to be classified

[0086] S42, inputting the test data into a pre-trained RecycleNet network;

[0087] S43, applying the softmax activation function in the output layer of the RecycleNet network to generate the classification probability of each category;

[0088] S44, according to the output probability of the softmax activation function, the category with the highest probability is determined as the classification result to achieve effective detection and classification of small targets on the sea surface.

[0089] Based on similar inventive concepts, an embodiment of the present invention further provides a computer storage medium storing a readable program, which can execute the above-mentioned sea surface target detection and classification method based on RecycleNet when the program is running.

[0090] Based on similar inventive concepts, an embodiment of the present invention provides an electronic device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus;

[0091] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above-mentioned sea surface target detection and classification method based on RecycleNet.

[0092] Based on similar inventive concepts, an embodiment of the present invention further provides a computer program product, including computer instructions, which instruct a computing device to perform operations corresponding to the above-mentioned sea surface target detection and classification method based on RecycleNet.

[0093] Example 2

[0094] In this embodiment, the effect of the method of the present invention is verified by a specific experiment:

[0095] In this embodiment, the data measured by Professor S. Haykin of McMaster University in Canada using IPIX radar in a real ocean environment is used as the radar to obtain the sea surface echo amplitude sequence of this embodiment.

[0096] In the experiment, the CommonNet network is used as a comparative experiment. The RecycleNet framework diagram of the method of the present invention is as follows: Figure 2 As shown, the F1 score values ​​of RecycleNet and CommonNet networks are shown in Table 1:

[0097] Table 1 F1 score values ​​of ten groups of data

[0098]

[0099] As can be seen from Table 1, the F1 score of RecycleNet of the method proposed in the present invention can reach above 0.9, which is much larger than CommonNet. This shows that RecycleNet performs better in precision and recall, and has better overall performance. This is because RecycleNet can recollect lost information, thereby improving the effectiveness and reliability of the classification task. A low loss value means that the network can better fit the data and reduce errors during training, thereby improving the overall prediction performance. Therefore, the RecycleNet network proposed in the present invention is superior to traditional classification methods in classification performance, improves efficiency, and accelerates the entire processing process.

[0100] Example 3

[0101] Based on the sea surface target detection and classification method based on RecycleNet proposed in Example 1, this embodiment proposes a sea surface target detection and classification system based on RecycleNet, such as Figure 3 As shown, specifically including:

[0102] Radar signal receiving and processing module: Use radar to obtain the sea surface echo amplitude sequence, and use recursive graph construction technology to process and obtain recursive graph data;

[0103] Network building module: building the RecycleNet network;

[0104] Network training module: using the recursive graph data to train the RecycleNet network and adjusting the network parameters to improve the classification accuracy of small targets;

[0105] Classification module: Use the trained RecycleNet network to perform classification and obtain classification results.

[0106] The method of the present invention may be implemented in hardware, firmware, or as software or computer code that may be stored in a recording medium (such as a CDROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code that is originally stored in a remote recording medium or a non-temporary machine-readable medium downloaded over a network and will be stored in a local recording medium, so that the method described herein may be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that a computer, processor, microprocessor controller, or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code, and when the software or computer code is accessed and executed by a computer, processor, or hardware, the method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the method shown herein, the execution of the code converts the general-purpose computer into a dedicated computer for executing the method shown herein.

[0107] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A sea surface target detection and classification method based on RecycleNet, characterized in that: The following steps are involved: The sea surface echo amplitude sequence is obtained by radar, and the recursive graph construction technology is used to process the recursive graph data; Build the RecycleNet network; Using the recursive graph data to train the RecycleNet network and adjust network parameters; Use the trained RecycleNet network for classification to obtain the classification results.

2. According to claim 1, a method for detecting and classifying sea surface targets based on RecycleNet is characterized in that: The sea surface echo amplitude sequence is composed of N continuous pulse echo data received by the radar, forming a time series data set, and each pulse echo data contains its corresponding amplitude information.

3. The method for sea surface target detection and classification based on RecycleNet according to claim 1, characterized in that: The RecycleNet network includes multiple Recycle modules and a fully connected module; the Recycle module includes a convolutional layer, a Leaky ReLU activation layer, a maximum pooling layer and a subtraction layer for feature recycling; multiple Recycle modules are sequentially connected layer by layer to capture detailed features in the recursive graph through feature extraction and recycling mechanisms.

4. The method for detecting and classifying sea surface targets based on RecycleNet according to claim 3, characterized in that: A fully connected layer is set after each Recycle module to collect the recycling feature information θ of each Recycle module. k (X), k = 1, 2, ... K are sent to the fully connected layer to merge the information. The weight matrix of the fully connected layer is W k , the bias term is b k , then the output formula through the fully connected layer is: y k =W k θ k (X)+b k ,k=1,2,...K The feature θ after each recycling k (X) The output information after the fully connected layer is merged, and the merged output comprehensive feature map y can be expressed as: Where K is the total number of outputted recovery feature information.

5. The method for detecting and classifying sea surface targets based on RecycleNet according to claim 4, characterized in that: The final module of the RecycleNet network is a fully connected module, which receives the comprehensive feature map y and maps it to the classification space; the output of the fully connected module is a score, for each category i, let the score be y i ; The output of the fully connected module is converted into probability values ​​of each category through the softmax activation function. Assuming that the total number of categories is C, the probability of the i-th category is: in, represents the score of the i-th category.

6. A method for detecting and classifying sea surface targets based on RecycleNet according to claim 1 or 5, characterized in that: The steps for classification using the RecycleNet network include: S41, preparing test data to be classified; S42, inputting the test data into the trained RecycleNet network; S43, applying a softmax activation function to the output of the RecycleNet network to generate classification probabilities for each category; S44, according to the output probability of the softmax activation function, the category with the highest probability is determined as the classification result.

7. A sea surface target detection and classification system based on RecycleNet, characterized in that: include: Radar signal receiving and processing module: Use radar to obtain the sea surface echo amplitude sequence, and use recursive graph construction technology to process and obtain recursive graph data; Network building module: building the RecycleNet network; Network training module: using the recursive graph data to train the RecycleNet network and adjust network parameters; And, classification module: use the trained RecycleNet network to perform classification and obtain the classification results.

8. A computer storage medium storing a readable program, characterized in that: When the program is running, it can execute the sea surface target detection and classification method based on RecycleNet as described in any one of claims 1 to 6.

9. An electronic device, characterized in that: include: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to a sea surface target detection and classification method based on RecycleNet as described in any one of claims 1-6.

10. A computer program product comprising computer instructions, characterized in that: The computer instructions instruct the computing device to perform operations corresponding to the sea surface target detection and classification method based on RecycleNet as described in any one of claims 1-6.

Images