High-precision radar target recognition system based on orthogonal waveform design and convolutional neural network

Abstract

The invention discloses a high-precision radar target recognition system based on orthogonal waveform design and a convolutional neural network. The high-precision radar target recognition system comprises a radar orthogonal waveform design module, a waveform generation module, a transmitting antenna module, a receiving antenna module, a waveform receiving module and a convolutional neural network module, by designing an orthogonal radar transmitting pulse string, the receiving entropy is increased, so that a radar echo pulse string received by a receiving antenna contains enough target information, and strong difference is presented between a target and clutter as well as multipath interference characteristics; therefore, the convolutional neural network can extract target key features from a radar echo pulse string received by an antenna and suppress clutter and multipath interference at the same time. According to the method, radar echoes are directly processed in the electromagnetic original domain, features are efficiently extracted, high-precision recognition is achieved, traditional multi-step signal processing is integrated into one step, the process is simplified, the hardware complexity of a receiver can be reduced possibly, and the method has great significance in improving the performance of an electronic information system.

Description

Technical Field

[0001] The present invention relates to radar target recognition technology and deep learning technology, and in particular to a high-precision radar target recognition system based on orthogonal waveform design and convolutional neural network. Background Art

[0002] Radar technology can detect targets all day, all weather and at long distances, becoming an important tool for target recognition. Convolutional neural networks have been widely used in target recognition tasks due to their excellent feature extraction capabilities. By processing the radar echo, carriers carrying target information such as one-dimensional images and two-dimensional images are obtained, and then the neural network is used to extract and identify the target information carrier to output the target type. For example, Jinhu Li, Fangzheng Zhang, Yu Xiang, and Shilong Pan, "Towards small target recognition with photonics-based highresolution radar range profiles," Opt. Express 29, 31574-31581, 2021. A photonic de-slant radar system is used to transmit a signal with a bandwidth of 8 GHz to detect the target, and the received radar echo is pulse compressed to obtain a one-dimensional image, and then the convolutional neural network is used to extract and identify the one-dimensional image, achieving a target recognition accuracy of up to 97.16%. Anyi Deng, Na Qian, Shiyu Hua, Jun Wan, Zhenbin Lv, and Weiwen Zou,"High-resolution ISAR imaging based on photonic receiving for high-accuracyautomatic target recognition," Opt. Express 30, 20580-20588, 2022 used a photon RF direct sampling radar system to transmit a signal with a bandwidth of 8 GHz to detect complex targets, and then performed pulse compression and coherent accumulation on multiple radar echoes after receiving to obtain a two-dimensional image. Then, a convolutional neural network was used to extract features and identify the two-dimensional image, achieving a recognition accuracy of 95%. However, these studies were all conducted in an ideal environment without clutter and multipath interference.

[0003] As the electromagnetic environment becomes increasingly complex, radar echoes are inevitably mixed with clutter and multipath interference in addition to target scattered echoes. These interference factors form false targets in the target information carrier, which seriously affects the target recognition results. Therefore, before the target information carrier is input into the convolutional neural network, effective preprocessing must be performed to remove false targets. However, these preprocessing algorithms are not only highly complex, but also applicable to limited types of clutter or interference (M.Azimifar, and M.A.Sebt, Clutter Suppression in Passive ISAR using Compressive Sensing, in IEEE Transactions on Aerospace and Electronic Systems, 2024).

[0004] Therefore, it is of great significance to study a simpler, more efficient and more accurate radar target recognition method in clutter and multipath interference environments. Summary of the invention

[0005] The purpose of the present invention is to address the deficiencies of the prior art and propose a high-precision radar target recognition system based on orthogonal waveform design and convolutional neural network. By designing the radar transmit pulse train as an orthogonal signal set, a strong difference between the target characteristics and the clutter and multipath interference characteristics is constructed, so that the convolutional neural network can extract the key characteristics of the target from the radar echo pulse train contaminated by clutter and multipath interference and suppress clutter and multipath interference, thereby achieving high-precision target recognition.

[0006] The present invention is implemented by adopting the following technical solutions.

[0007] A high-precision radar target recognition system based on orthogonal waveform design and convolutional neural network, which has the following characteristics:

[0008] Radar orthogonal waveform design module, used to design orthogonal signal sets to increase the information entropy of radar transmission signals;

[0009] A waveform generation module, connected to the radar orthogonal waveform design module, for generating an orthogonal pulse train designed by the radar orthogonal waveform design module;

[0010] A transmitting antenna module, connected to the waveform generating module, and used to transmit the orthogonal pulse train as an electromagnetic signal;

[0011] A receiving antenna module, used for receiving radar echo pulse trains;

[0012] A waveform receiving module, connected to the receiving antenna module, for converting a received radar echo pulse train into a digital signal;

[0013] A convolutional neural network module, connected to the waveform receiving module, for identifying the target type based on the converted radar echo pulse train digital signal;

[0014] Among them, the orthogonal pulse train signal designed by the radar orthogonal waveform design module can be scattered on the surface of the object after being transmitted by the waveform generation module and the antenna transmission module, and is received by the antenna receiving module and the waveform receiving module to form a data set of radar echo pulse trains. The data set contains sufficient target information, and the target characteristics are highly different from the clutter and multipath interference characteristics for the convolutional neural network module; by training the convolutional neural network module, the convolutional neural network module can learn and extract key target features from the data set, while suppressing clutter and multipath interference, thereby achieving high-precision target recognition.

[0015] The method for realizing target recognition by using the above system comprises the following steps:

[0016] 1) Training phase:

[0017] During the training phase, it is necessary to adjust the orthogonal pulse train designed by the radar orthogonal waveform design module to increase the emission entropy; design the hyperparameters such as the convolutional neural network architecture; use the optimization algorithm to efficiently train the convolutional neural network; use the data set constructed by the radar echo pulse train received by the receiving antenna module to train the convolutional neural network; when the convolutional neural network target recognition accuracy is low (such as <90%), adjust the radar orthogonal waveform design module and the convolutional neural network module, so that the convolutional neural network can finally extract the key features of the target from the radar echo pulse train polluted by clutter and multipath interference and suppress clutter and multipath interference, and recognize the target based on the radar echo with high precision. Fix the parameters of the radar orthogonal waveform design module and the convolutional neural network module.

[0018] 2) Application phase:

[0019] Based on the orthogonal pulse string designed by the radar orthogonal waveform design module in the training phase and the parameters of the convolutional neural network module, in the application phase, the transmitting antenna module is used to transmit the orthogonal pulse string signal designed by the radar orthogonal waveform design module generated by the signal generating module, and the antenna receiving module receives the radar echo pulse string, which serves as the input of the convolutional neural network, and the convolutional neural network outputs the target type.

[0020] Compared with the prior art, the present invention has the following advantages:

[0021] 1. The present invention designs the radar transmit pulse train as an orthogonal signal set to increase the information entropy of the transmit signal, that is, the transmit entropy, and constructs a strong difference between the target characteristics and the clutter and multipath interference characteristics, so that the convolutional neural network can correctly identify the target in the presence of clutter and multipath interference. The present invention combines multiple steps such as pulse compression, clutter and multipath interference suppression, and feature extraction in the existing radar target recognition into one step, thereby simplifying the processing steps.

[0022] 2. The present invention trains a convolutional neural network to automatically learn target features and suppress clutter and multipath interference from a data set with strong differences. Compared with traditional clutter and multipath interference suppression methods, it is simpler, more efficient and more universal.

[0023] 3) The radar orthogonal waveform design module significantly increases the information entropy (i.e., emission entropy) of the radar transmit signal by designing an orthogonal signal set. This design method can enhance the uniqueness and distinguishability of the signal, allowing the radar system to more accurately distinguish different targets, thereby improving the accuracy of target recognition. The convolutional neural network module can extract the key features of the target from the radar echo pulse train through deep learning and feature extraction technology, and effectively suppress clutter and multipath interference. This capability enables the system to accurately identify targets in complex environments, further improving the accuracy of recognition.

[0024] 4) Convolutional neural networks have low computational complexity and spatial complexity, and are suitable for large-scale data processing and real-time application scenarios. In radar target recognition systems, this feature enables the system to quickly process a large amount of radar echo data and achieve real-time target detection and recognition. By optimizing the radar orthogonal waveform design and convolutional neural network architecture, the system can achieve rapid response and real-time application while ensuring recognition accuracy. This is of great significance to the military, aviation, transportation and other fields, and can provide decision makers with timely and accurate target information. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A schematic diagram of an embodiment of a high-precision radar target recognition system based on radar orthogonal waveform design and convolutional neural network of the present invention

[0026] Figure 2 Schematic diagram of an embodiment of a radar orthogonal pulse train, wherein a shows the Welch Costas sequence encoding, and b shows the time-frequency diagram of 10 orthogonal signals combining the Costas frequency hopping signal with a broadband linear frequency modulation (LFM) signal;

[0027] Figure 3 A schematic diagram of an embodiment of a convolutional neural network architecture

[0028] Figure 4Target feature distribution diagram under different radar emission waveforms in one embodiment DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be described clearly and completely in combination with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all 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.

[0030] This implementation provides a radar target recognition system, which achieves high-precision recognition of targets in clutter and multipath interference environments by designing the orthogonal pulse train waveform emitted by the radar and combining it with the powerful feature extraction capability of convolutional neural networks. Figure 1 The block diagram of an embodiment of the system of the present invention is shown, which mainly includes the following radar orthogonal waveform design module 1, waveform generation module 2, transmitting antenna module 3, receiving antenna module 4, waveform receiving module 5 and convolutional neural network module 6. The output end of the waveform generation module 2 is connected to the input end of the transmitting antenna module 3, the output end of the receiving antenna module 4 is connected to the input end of the waveform receiving module 5, and the output end of the waveform receiving module 5 is connected to the input end of the convolutional neural network module 6. Among them, the radar waveform design module 1 is responsible for designing the orthogonal radar transmission pulse train waveform to ensure that the waveform information entropy is large enough to enhance the characteristic difference between the target and clutter and multipath interference. The above-mentioned waveform generation module 2 is an arbitrary waveform generator, which can generate corresponding orthogonal waveform signals according to the output of the radar orthogonal waveform design module. The transmitting antenna module 3 includes a multi-band antenna for transmitting the electromagnetic signal generated by the waveform generation module 2. The receiving antenna array module 4 is also equipped with a multi-band antenna for receiving radar echo pulse trains to ensure stable reception of signals within a wide spectrum range. The waveform receiving module 5 is responsible for receiving and converting the radar echo captured by the antenna receiving module 4 into a digital signal. The waveform receiving module can be an electrical system or an optoelectronic hybrid system. The above-mentioned convolutional neural network 6 is implemented on a computer to extract the key features of the target from the radar echo pulse train received by the receiving module, and suppress clutter and multipath interference to achieve high-precision target identification.

[0031] The above-mentioned high-precision radar target recognition implementation method based on radar orthogonal waveform design and convolutional neural network is divided into two stages: training stage and application stage, which are described in detail as follows.

[0032] Training phase:

[0033] First waveform design: Use the radar orthogonal waveform design module 1 to design an orthogonal pulse train waveform so that the radar echo pulse train received by the receiving antenna contains sufficient target information and there is a strong difference between the target and clutter and multipath interference characteristics, so that the convolutional neural network module can extract the key features of the target from the radar echo pulse train received by the antenna while suppressing clutter and multipath interference. Figure 2 FIG. 1 is a schematic diagram of an embodiment of a radar orthogonal pulse train. Figure 2 a shows the Welch Costas sequence encoding, Figure 2 b shows the time-frequency diagram of 10 orthogonal signals that are a combination of the Costas frequency hopping signal and a broadband linear frequency modulation (LFM) signal, wherein the LFM slopes of the 10 orthogonal signals alternate between positive and negative.

[0034] Generation, transmission and reception of the second orthogonal pulse train signal: the waveform generation module 2 is used to generate the pulse train signal, the transmitting antenna module 3 transmits the pulse train signal, and the receiving antenna module 4 and the waveform receiving module 5 receive the radar echo pulse train.

[0035] The third data set is constructed by using the radar echo pulse train to construct a data set to train the convolutional neural network module 6.

[0036] Fourth, optimize the convolutional neural network architecture, hyperparameters, training methods, etc. to verify whether the network can achieve high-precision target recognition. Figure 3 As shown, this embodiment includes an input layer, 4 convolutional layers and 4 maximum pooling layers, 2 fully connected layers and an output layer.

[0037] Fifth, when the optimized convolutional neural network still cannot achieve ideal output, optimize the radar orthogonal waveform design module 1 so that after repeating steps one to four above, the convolutional neural network module 6 can achieve high-precision target recognition. Figure 4 is a target feature distribution diagram under different radar emission waveforms in one embodiment. Figure 4 It can be seen from a that the convolutional neural network can more easily learn the key features of the target from the data set constructed by the orthogonal radar echo pulse train. The two types of target features have obvious linear separability and are easy to be correctly identified.

[0038] Application phase:

[0039] The orthogonal pulse train obtained by the radar orthogonal waveform design module 1 in the training stage and the parameters of the convolutional neural network module 6 are used. The waveform generation module 2 and the transmitting antenna module 3 are used to generate the designed radar pulse train and transmit it, the receiving antenna module 4 receives the radar echo pulse train, and the waveform receiving module 6 converts the radar echo pulse train into data or signals adapted to the input of the convolutional neural network module 6. The convolutional neural network module 6 extracts the key features of the target from the radar echo pulse train and suppresses clutter and multipath interference, so as to identify the target type with high precision.

[0040] The present invention processes radar echoes in the electromagnetic original domain based on radar orthogonal waveform design and convolutional neural network, and efficiently extracts target features from them. It combines the pulse compression, clutter interference suppression, feature extraction and other multi-step processes of traditional radar target recognition into one step, simplifies the processing flow and even reduces the complexity of the receiver hardware system. The present invention can be widely used in the field of target recognition.

Claims

1. A high-precision radar target recognition system based on orthogonal waveform design and convolutional neural network, characterized in that: include: A radar orthogonal waveform design module (1), used for designing an orthogonal signal set to increase the information entropy of the radar transmission signal; A waveform generating module (2), connected to the radar orthogonal waveform design module, for generating an orthogonal pulse train designed by the radar orthogonal waveform design module; A transmitting antenna module (3), connected to the waveform generating module, and used for transmitting the orthogonal pulse train as an electromagnetic signal; A receiving antenna module (4), used for receiving a radar echo pulse train; A waveform receiving module (5), connected to the receiving antenna module, for converting a received radar echo pulse train into a digital signal; A convolutional neural network module (6), connected to the waveform receiving module, for identifying the target type based on the converted radar echo pulse train digital signal; The orthogonal pulse train signal designed by the radar orthogonal waveform design module (1) can be scattered on the surface of an object after being transmitted by the waveform generation module (2) and the antenna transmission module (3), and is received by the antenna receiving module (4) and the waveform receiving module (5) to form a data set of radar echo pulse trains, wherein the data set contains sufficient target information, and the target characteristics have strong differences from the clutter and multipath interference characteristics, so as to be provided to the convolutional neural network module (6); by training the convolutional neural network module (6), the convolutional neural network module (6) can learn and extract key target features from the data set, while suppressing clutter and multipath interference, thereby achieving high-precision target recognition.

2. According to claim 1, the high-precision radar target recognition technology based on orthogonal waveform design and convolutional neural network is characterized in that: The training of the convolutional neural network module (6) comprises: Adjust the radar orthogonal waveform design module and design orthogonal radar pulse trains to increase the transmission entropy; Design and optimize convolutional neural network architecture and hyperparameters; The convolutional neural network is trained using a dataset constructed using radar echo pulse trains until the recognition accuracy requirements are met.

3. The high-precision radar target recognition technology based on orthogonal waveform design and convolutional neural network according to claim 2 is characterized in that: The waveform generation module (2) is capable of generating any orthogonal waveform sequence within a wide spectrum range.

4. According to claim 2, the high-precision radar target recognition technology based on orthogonal waveform design and convolutional neural network is characterized in that: The transmitting antenna module (3) can realize signal transmission within a wide spectrum range.

5. The high-precision radar target recognition technology based on orthogonal waveform design and convolutional neural network according to claim 2 is characterized in that: The receiving antenna module (4) can realize signal reception within a wide spectrum range.

6. The high-precision radar target recognition technology based on orthogonal waveform design and convolutional neural network according to claim 2 is characterized in that: The waveform receiving module (5) is a device capable of converting an orthogonal radar echo pulse train received by an antenna into a digital signal.

Images