Submit Manuscript
Peer Review
Editor Work2023 Vol. 12, No. 6
Previous Issue
2023, 12(6): 1125-1137.
Abstract
1504KB
Forward-looking imaging is crucial in many civil and military fields, such as precision guidance, autonomous landing, and autonomous driving. The forward-looking imaging performance of airborne radar may deteriorate significantly due to the constraint of the Doppler history. The deconvolution method can be used to improve the quality of forward-looking imaging; however, it will not work well for complex imaging scenes. To solve the problem of scene sparsity measurement and characterization in complex forward-looking imaging configurations, an efficient probability model-driven airborne Bayesian forward-looking super-resolution imaging algorithm is proposed for multitarget scenarios to improve the azimuth resolution. First, the data dimension of the forward-looking imaging scene was expanded from single-frame to multiframe spaces to enhance the sparsity of the imaging scene. Then, the sparse characteristics of the imaging scene were statistically modeled using the generalized Gaussian probability model. Finally, the super-resolution imaging problem was solved using the Bayesian framework. Because the sparsity characterization parameters are embedded in the entire process of imaging, the forward-looking imaging parameters will be updated during each iteration. The effectiveness of the proposed algorithm was verified using simulation and real data.
Forward-looking imaging is crucial in many civil and military fields, such as precision guidance, autonomous landing, and autonomous driving. The forward-looking imaging performance of airborne radar may deteriorate significantly due to the constraint of the Doppler history. The deconvolution method can be used to improve the quality of forward-looking imaging; however, it will not work well for complex imaging scenes. To solve the problem of scene sparsity measurement and characterization in complex forward-looking imaging configurations, an efficient probability model-driven airborne Bayesian forward-looking super-resolution imaging algorithm is proposed for multitarget scenarios to improve the azimuth resolution. First, the data dimension of the forward-looking imaging scene was expanded from single-frame to multiframe spaces to enhance the sparsity of the imaging scene. Then, the sparse characteristics of the imaging scene were statistically modeled using the generalized Gaussian probability model. Finally, the super-resolution imaging problem was solved using the Bayesian framework. Because the sparsity characterization parameters are embedded in the entire process of imaging, the forward-looking imaging parameters will be updated during each iteration. The effectiveness of the proposed algorithm was verified using simulation and real data.
2023, 12(6): 1138-1154.
Distinguishing multiple targets in the same resolution cell is an important and challenging task in the forward-looking imaging process of monopulse radar. Although Doppler processing can improve the recognition performance for multiple targets at high squint angles, the precise estimation of Doppler frequency remains challenging under conditions with unknown target numbers and energy leakage from strong point targets. To address these issues, this paper proposes a Fast Iterative Interpolated Beamforming (FIIB) algorithm with model order estimation and single snapshot processing for monopulse forward-looking imaging, which combines information theory to unbiasedly estimate the number of targets and Doppler frequencies. The simulation results show the superiority of the proposed FIIB algorithm over the Chirp-Z Transform (CZT) algorithm for estimating target numbers and Doppler frequencies within the same resolution cell in the presence of multiple point targets. In addition, the proposed FIIB algorithm can accurately estimate point targets beyond a ±5° azimuth angle in monopulse angle measurement tasks. Real-data experiments also reveal that FIIB-based monopulse forward-looking imaging has high focusing capability and imaging contrast and can effectively suppress background clutter.
Distinguishing multiple targets in the same resolution cell is an important and challenging task in the forward-looking imaging process of monopulse radar. Although Doppler processing can improve the recognition performance for multiple targets at high squint angles, the precise estimation of Doppler frequency remains challenging under conditions with unknown target numbers and energy leakage from strong point targets. To address these issues, this paper proposes a Fast Iterative Interpolated Beamforming (FIIB) algorithm with model order estimation and single snapshot processing for monopulse forward-looking imaging, which combines information theory to unbiasedly estimate the number of targets and Doppler frequencies. The simulation results show the superiority of the proposed FIIB algorithm over the Chirp-Z Transform (CZT) algorithm for estimating target numbers and Doppler frequencies within the same resolution cell in the presence of multiple point targets. In addition, the proposed FIIB algorithm can accurately estimate point targets beyond a ±5° azimuth angle in monopulse angle measurement tasks. Real-data experiments also reveal that FIIB-based monopulse forward-looking imaging has high focusing capability and imaging contrast and can effectively suppress background clutter.
Forward-looking Multi-channel Synthetic Aperture Radar Imaging and Array Attitude Error Compensation
2023, 12(6): 1155-1165.
To solve the problem of left-right Doppler ambiguity in forward-looking Synthetic Aperture Radar (SAR) imaging, Forward-Looking Multi-Channel SAR (FLMC-SAR) can achieve Doppler ambiguity resolving imaging through beamforming. However, array deviation angle and time-varying platform attitude errors lead to the mismatch between the space and time characteristics of a target, which affects the Doppler ambiguity resolving imaging performance. This study proposes an FLMC-SAR imaging and array attitude error compensation method. First, the three-dimensional FLMC-SAR array deviation-angle error and time-varying platform attitude error models are established. The mechanism of matching the spatiotemporal characteristics of targets in the two-dimensional spatiotemporal spectrum plane is analyzed. In addition, a representation model is established to study the spatiotemporal characteristics mismatch caused by the array attitude error in the spatiotemporal plane. Thereafter, based on the non-left-right spatial variation of the error, we propose a method of adding an error compensation phase to the back projection function to uniformly compensate the array attitude error of left-right symmetric targets. The results of simulation experiments show that the proposed method can achieve FLMC-SAR array attitude correction and error compensation, enhance the performance of forward-looking Doppler ambiguity suppression, and ensure the azimuth resolution performance of forward-looking imaging.
To solve the problem of left-right Doppler ambiguity in forward-looking Synthetic Aperture Radar (SAR) imaging, Forward-Looking Multi-Channel SAR (FLMC-SAR) can achieve Doppler ambiguity resolving imaging through beamforming. However, array deviation angle and time-varying platform attitude errors lead to the mismatch between the space and time characteristics of a target, which affects the Doppler ambiguity resolving imaging performance. This study proposes an FLMC-SAR imaging and array attitude error compensation method. First, the three-dimensional FLMC-SAR array deviation-angle error and time-varying platform attitude error models are established. The mechanism of matching the spatiotemporal characteristics of targets in the two-dimensional spatiotemporal spectrum plane is analyzed. In addition, a representation model is established to study the spatiotemporal characteristics mismatch caused by the array attitude error in the spatiotemporal plane. Thereafter, based on the non-left-right spatial variation of the error, we propose a method of adding an error compensation phase to the back projection function to uniformly compensate the array attitude error of left-right symmetric targets. The results of simulation experiments show that the proposed method can achieve FLMC-SAR array attitude correction and error compensation, enhance the performance of forward-looking Doppler ambiguity suppression, and ensure the azimuth resolution performance of forward-looking imaging.
2023, 12(6): 1166-1178.
The direction of arrival estimation algorithm can overcome the Rayleigh limit, effectively separate multiple targets within the main lobe, and improve the azimuth resolution when applied to forward-looking imaging in airborne multi-channel radar. However, the limited antenna beam coverage and rapid scanning result in a scarcity of data samples available for covariance matrix estimation, leading to direction and amplitude estimation errors. Herein, we propose a forward-looking imaging algorithm based on single-snapshot iterative super-resolution estimation. The algorithm performs single-snapshot iterative spectral estimation to accurately determine the azimuth and amplitude of the target. Subsequently, a high-resolution image is achieved through non-coherent accumulation. Simulation and experimental data processing results show that the proposed algorithm can resolve multiple targets, significantly improving the azimuth resolution of the forward-looking image compared with traditional forward-looking imaging algorithms. Moreover, it ensures the accurate reconstruction of point targets and contour reconstruction of area targets.
The direction of arrival estimation algorithm can overcome the Rayleigh limit, effectively separate multiple targets within the main lobe, and improve the azimuth resolution when applied to forward-looking imaging in airborne multi-channel radar. However, the limited antenna beam coverage and rapid scanning result in a scarcity of data samples available for covariance matrix estimation, leading to direction and amplitude estimation errors. Herein, we propose a forward-looking imaging algorithm based on single-snapshot iterative super-resolution estimation. The algorithm performs single-snapshot iterative spectral estimation to accurately determine the azimuth and amplitude of the target. Subsequently, a high-resolution image is achieved through non-coherent accumulation. Simulation and experimental data processing results show that the proposed algorithm can resolve multiple targets, significantly improving the azimuth resolution of the forward-looking image compared with traditional forward-looking imaging algorithms. Moreover, it ensures the accurate reconstruction of point targets and contour reconstruction of area targets.
2023, 12(6): 1179-1201.
Airborne Wide angle staring Synthetic Aperture Radar (WasSAR) is a novel SAR imaging technique that enables multiangle and long-time staring imaging of an observation region. The combination of WasSAR and Ground Moving Target Indication (GMTI) facilitates the long-time tracking of moving targets in key areas, thereby obtaining dynamic sensing information. Herein, we initially constructed a moving target echo model for airborne multi-channel WasSAR, followed by an analysis of the characteristics of a moving target during WasSAR imaging. Further, group phase shift calibration and modified A2DC methods are proposed to mitigate the influence of platform attitude errors and channel imbalances. In accordance with this, an extended airborne multi-channel WasSAR ground moving target detection and tracking method is proposed, which achieves accurate detection and tracking of targets moving on complex roads. Finally, a trajectory reconstruction method for moving targets in airborne multi-channel WasSAR-GMTI is presented, demonstrating accurate trajectory reconstruction for targets on complex roads. Moreover, a flight experiment using our independently developed airborne multi-channel WasSAR-GMTI system, along with the associated real data processing results, is presented. These results confirm the validity and practicability of airborne WasSAR-GMTI for moving target dynamic surveillance and serve as a foundation for future research.
Airborne Wide angle staring Synthetic Aperture Radar (WasSAR) is a novel SAR imaging technique that enables multiangle and long-time staring imaging of an observation region. The combination of WasSAR and Ground Moving Target Indication (GMTI) facilitates the long-time tracking of moving targets in key areas, thereby obtaining dynamic sensing information. Herein, we initially constructed a moving target echo model for airborne multi-channel WasSAR, followed by an analysis of the characteristics of a moving target during WasSAR imaging. Further, group phase shift calibration and modified A2DC methods are proposed to mitigate the influence of platform attitude errors and channel imbalances. In accordance with this, an extended airborne multi-channel WasSAR ground moving target detection and tracking method is proposed, which achieves accurate detection and tracking of targets moving on complex roads. Finally, a trajectory reconstruction method for moving targets in airborne multi-channel WasSAR-GMTI is presented, demonstrating accurate trajectory reconstruction for targets on complex roads. Moreover, a flight experiment using our independently developed airborne multi-channel WasSAR-GMTI system, along with the associated real data processing results, is presented. These results confirm the validity and practicability of airborne WasSAR-GMTI for moving target dynamic surveillance and serve as a foundation for future research.
2023, 12(6): 1202-1214.
The spaceborne-missile bistatic forward-looking Synthetic Aperture Radar (SAR) is a promising imaging guidance technology that can obtain high-resolution images of the area in front of the missile all day and in all weather types. However, the coupling and spatial variations in range and azimuth parameters hinder the development of high-resolution spaceborne-missile bistatic forward-looking SAR imaging. In this study, the accurate-range Doppler domain analytical formula for echo signals was derived based on the low-orbit spaceborne illuminator and high-speed forward-looking missile-borne receiving platform configuration. Subsequently, in range processing, a range Nonlinear Chirp Scaling (NCS) was proposed to equalize the range cell migration and range Frequency Modulation (FM) rate, and both can be uniformly compensated in the two-dimensional frequency domain. In azimuth processing, the proposed method decomposed the azimuth FM rates of the transmitter and receiver. Then, the azimuth NCS was used to eliminate the high-order spatial variation of the azimuth FM rate. Finally, a two-dimensional matched filtering was performed to obtain a SAR image with a good global focus. The point and scene simulation verify the effectiveness of the proposed algorithm.
The spaceborne-missile bistatic forward-looking Synthetic Aperture Radar (SAR) is a promising imaging guidance technology that can obtain high-resolution images of the area in front of the missile all day and in all weather types. However, the coupling and spatial variations in range and azimuth parameters hinder the development of high-resolution spaceborne-missile bistatic forward-looking SAR imaging. In this study, the accurate-range Doppler domain analytical formula for echo signals was derived based on the low-orbit spaceborne illuminator and high-speed forward-looking missile-borne receiving platform configuration. Subsequently, in range processing, a range Nonlinear Chirp Scaling (NCS) was proposed to equalize the range cell migration and range Frequency Modulation (FM) rate, and both can be uniformly compensated in the two-dimensional frequency domain. In azimuth processing, the proposed method decomposed the azimuth FM rates of the transmitter and receiver. Then, the azimuth NCS was used to eliminate the high-order spatial variation of the azimuth FM rate. Finally, a two-dimensional matched filtering was performed to obtain a SAR image with a good global focus. The point and scene simulation verify the effectiveness of the proposed algorithm.
2023, 12(6): 1215-1228.
Automatic Target Recognition (ATR) is an interdisciplinary technological field related to pattern recognition, artificial intelligence, and information processing. ATR evaluation focuses on accessing ATR algorithms and systems. Due to the noncooperative targets, complex operating conditions, and multiple subjective preferences of the decision maker, ATR evaluation is performed for the entire ATR research process and shows its importance in guiding ATR development. This paper presents the connotation of ATR evaluation and briefly reviews ATR development. Furthermore, the conventional methods, applications, and latest developments in ATR evaluation are presented and discussed from the perspective of performance measures, test condition, inference and decision. Finally, several ATR evaluation research directions are summarized. This paper serves as a valuable reference for a better understanding of ATR evaluation and the effective adoption of various ATR evaluation methods.
Automatic Target Recognition (ATR) is an interdisciplinary technological field related to pattern recognition, artificial intelligence, and information processing. ATR evaluation focuses on accessing ATR algorithms and systems. Due to the noncooperative targets, complex operating conditions, and multiple subjective preferences of the decision maker, ATR evaluation is performed for the entire ATR research process and shows its importance in guiding ATR development. This paper presents the connotation of ATR evaluation and briefly reviews ATR development. Furthermore, the conventional methods, applications, and latest developments in ATR evaluation are presented and discussed from the perspective of performance measures, test condition, inference and decision. Finally, several ATR evaluation research directions are summarized. This paper serves as a valuable reference for a better understanding of ATR evaluation and the effective adoption of various ATR evaluation methods.
2023, 12(6): 1229-1248.
Coherently combining distributed apertures adjusts the transmitted/received signals of multiple distributed small apertures, allowing coordinated distributed systems to obtain high power aperture products at much lower cost than large aperture. This is a promising and viable technology as an alternative to using large apertures. This study describes the concept and principles of coherently combining distributed apertures. Depending on whether external signal inputs at the combination destination are necessary, the implementation architecture of coherent combination is classified into two categories: closed- and open-loop. The development of coherently combining distributed apertures and their application in fields such as missile defense, deep space telemetry control, radar detection over ultralong range, and radio astronomy are then comprehensively presented. Furthermore, key techniques for aligning the time and phase of the transmitted/received signals for each aperture are elaborated, which are also necessary for coherently combining distributed apertures, including high-precision distributed time-frequency transfer and synchronization, and coherently combining parameters estimation, measurement and calibration, and prediction. Finally, summary is presented, and the scope of future works in this field is explored.
Coherently combining distributed apertures adjusts the transmitted/received signals of multiple distributed small apertures, allowing coordinated distributed systems to obtain high power aperture products at much lower cost than large aperture. This is a promising and viable technology as an alternative to using large apertures. This study describes the concept and principles of coherently combining distributed apertures. Depending on whether external signal inputs at the combination destination are necessary, the implementation architecture of coherent combination is classified into two categories: closed- and open-loop. The development of coherently combining distributed apertures and their application in fields such as missile defense, deep space telemetry control, radar detection over ultralong range, and radio astronomy are then comprehensively presented. Furthermore, key techniques for aligning the time and phase of the transmitted/received signals for each aperture are elaborated, which are also necessary for coherently combining distributed apertures, including high-precision distributed time-frequency transfer and synchronization, and coherently combining parameters estimation, measurement and calibration, and prediction. Finally, summary is presented, and the scope of future works in this field is explored.
Consistency-based Air Target Height Estimation and Location in Distributed Space-based Radar Network
2023, 12(6): 1249-1262.
When single space-based radar tracks and detects air targets, problems such as missing pitch angle information and nonlinear measurement lead to large target height estimation errors. Multi-space-based radar networking can solve this problem. Moreover, considering the system’s requirements for low computational complexity, low communication overhead, high accuracy, and high reliability, a consistency-based method for height estimation and location of air targets in a distributed space-based radar network is proposed. First, an air target motion model and a space-based radar measurement model are presented. Second, based on probabilistic graphical model theory, a factor graph for multi-frame measurement of target tracking and positioning in a space-based radar network is established. The coupling relationship between several local target motion states is established based on consistency fusion. Third, combining particle filtering and belief propagation establishes the message representation and iterative calculation rules of nonparametric belief propagation on the fusion tracking for factor graph of space-based radar networking. Finally, the performance of the algorithm is tested through simulation. The simulation results show that compared with the distributed consensus extended Kalman filter, the proposed algorithm improves the target height estimation accuracy by 35.3%, effectively improving the target localization performance of the space-based radar.
When single space-based radar tracks and detects air targets, problems such as missing pitch angle information and nonlinear measurement lead to large target height estimation errors. Multi-space-based radar networking can solve this problem. Moreover, considering the system’s requirements for low computational complexity, low communication overhead, high accuracy, and high reliability, a consistency-based method for height estimation and location of air targets in a distributed space-based radar network is proposed. First, an air target motion model and a space-based radar measurement model are presented. Second, based on probabilistic graphical model theory, a factor graph for multi-frame measurement of target tracking and positioning in a space-based radar network is established. The coupling relationship between several local target motion states is established based on consistency fusion. Third, combining particle filtering and belief propagation establishes the message representation and iterative calculation rules of nonparametric belief propagation on the fusion tracking for factor graph of space-based radar networking. Finally, the performance of the algorithm is tested through simulation. The simulation results show that compared with the distributed consensus extended Kalman filter, the proposed algorithm improves the target height estimation accuracy by 35.3%, effectively improving the target localization performance of the space-based radar.
2023, 12(6): 1263-1274.
Frequency agile technology provides full play to the advantage of radars for adopting electronic countermeasures actively, which can effectively enhance the antinoise suppression jamming performance of radars. However, with the increasing complexity of the interference environment, developing an online decision-making method for frequency-agile radar with dynamic adaptability and without foresight of the nature of the environment is a demanding task. According to the features of the jamming strategy, suppression jamming scenarios are divided into three categories, and an online decision-making method for frequency-agile radar based on Multi-Armed Bandit (MAB) is developed to maximize the radar’s detection probability. This approach is an online learning algorithm that does not need to interfere with the foresight of the environment and offline training process and realizes remarkable learning performance from noninterference scenarios to adaptive interference scenarios. The simulation results and theoretical analysis demonstrate that compared with the classical algorithm and stochastic agile strategy, the proposed method has stronger flexibility and can effectively improve the antijamming and target detection performances of the frequency-agile radar for various jamming scenarios.
Frequency agile technology provides full play to the advantage of radars for adopting electronic countermeasures actively, which can effectively enhance the antinoise suppression jamming performance of radars. However, with the increasing complexity of the interference environment, developing an online decision-making method for frequency-agile radar with dynamic adaptability and without foresight of the nature of the environment is a demanding task. According to the features of the jamming strategy, suppression jamming scenarios are divided into three categories, and an online decision-making method for frequency-agile radar based on Multi-Armed Bandit (MAB) is developed to maximize the radar’s detection probability. This approach is an online learning algorithm that does not need to interfere with the foresight of the environment and offline training process and realizes remarkable learning performance from noninterference scenarios to adaptive interference scenarios. The simulation results and theoretical analysis demonstrate that compared with the classical algorithm and stochastic agile strategy, the proposed method has stronger flexibility and can effectively improve the antijamming and target detection performances of the frequency-agile radar for various jamming scenarios.
2023, 12(6): 1275-1289.
Radio Frequency-screen is one of the earliest radar active antijamming measures. It achieves antijamming by transmitting cover pulses of different frequencies before the radar pulse signal to induce enemy jammers. As the demand for antijamming measures has become increasingly urgent in recent years, Radio Frequency-screen technology has been further developed. The most representative is the use of discontinuous spectrum signals as a cover signal. However, energy utilization for sending the cover signal can be improved further. To address this problem, this paper proposes a discrete spectrum cover signal based on the discontinuous spectrum cover signal and establishes the waveform design function under the joint constraint of constant modulus and spectral amplitude. The cover signal with discrete spectrum and energy aggregation is generated using the Alternating Direction Method of Multipliers (ADMM) and spectrum shaping algorithm solution. The simulation results show that the discrete spectrum cover signal has a higher spectral amplitude of approximately 5～12 dB than the discontinuous-spectrum cover signal for the same energy and bandwidth. Moreover, the discrete spectrum cover signal can cover a larger spectral range with the same energy and close spectral amplitude, realizing a better antijamming cover effect.
Radio Frequency-screen is one of the earliest radar active antijamming measures. It achieves antijamming by transmitting cover pulses of different frequencies before the radar pulse signal to induce enemy jammers. As the demand for antijamming measures has become increasingly urgent in recent years, Radio Frequency-screen technology has been further developed. The most representative is the use of discontinuous spectrum signals as a cover signal. However, energy utilization for sending the cover signal can be improved further. To address this problem, this paper proposes a discrete spectrum cover signal based on the discontinuous spectrum cover signal and establishes the waveform design function under the joint constraint of constant modulus and spectral amplitude. The cover signal with discrete spectrum and energy aggregation is generated using the Alternating Direction Method of Multipliers (ADMM) and spectrum shaping algorithm solution. The simulation results show that the discrete spectrum cover signal has a higher spectral amplitude of approximately 5～12 dB than the discontinuous-spectrum cover signal for the same energy and bandwidth. Moreover, the discrete spectrum cover signal can cover a larger spectral range with the same energy and close spectral amplitude, realizing a better antijamming cover effect.
2023, 12(6): 1290-1304.
In modern electronic warfare, the jamming environment of radar is more complex than ever. The airborne jammer adapts its jamming method based on diverse raid missions and stages. Recently, the reinforcement learning–based radar anti-jamming method has made some progress in the confrontation scenario of single jamming; however, the gap with respect to actual complex multi-jamming scenarios is large. To address this issue, this paper proposes a multi-jamming scenario radar anti-jamming method based on deep reinforcement learning in the complex domain to optimize the anti-jamming strategy of frequency agile radar. First, according to the stage characteristics of the raid mission, noise spot jamming, range deception jamming , and dense false target forwarding jamming models are established. The three jamming sequence strategies were designed to simulate actual jamming scenarios. Second, a reinforcement learning reward function that integrates the signal-to-noise ratio and target trajectory integrity is constructed for the multi-jamming scenario model. Thus, a multi-jamming scenario radar anti-jamming method based on deep reinforcement learning in a complex domain is proposed, which is based on the complex domain characteristics of the jamming signal. Finally, radar anti-jamming simulation experiments are performed based on the three jamming sequence strategies. The results show that the proposed method can effectively deal with the main-lobe jamming problem of complex multi-jamming scenarios under time-sequence conditions. Moreover, the average decision-making accuracy was improved, and the average decision-making time was reduced to 405.3 ms compared with the two classical reinforcement learning algorithms.
In modern electronic warfare, the jamming environment of radar is more complex than ever. The airborne jammer adapts its jamming method based on diverse raid missions and stages. Recently, the reinforcement learning–based radar anti-jamming method has made some progress in the confrontation scenario of single jamming; however, the gap with respect to actual complex multi-jamming scenarios is large. To address this issue, this paper proposes a multi-jamming scenario radar anti-jamming method based on deep reinforcement learning in the complex domain to optimize the anti-jamming strategy of frequency agile radar. First, according to the stage characteristics of the raid mission, noise spot jamming, range deception jamming , and dense false target forwarding jamming models are established. The three jamming sequence strategies were designed to simulate actual jamming scenarios. Second, a reinforcement learning reward function that integrates the signal-to-noise ratio and target trajectory integrity is constructed for the multi-jamming scenario model. Thus, a multi-jamming scenario radar anti-jamming method based on deep reinforcement learning in a complex domain is proposed, which is based on the complex domain characteristics of the jamming signal. Finally, radar anti-jamming simulation experiments are performed based on the three jamming sequence strategies. The results show that the proposed method can effectively deal with the main-lobe jamming problem of complex multi-jamming scenarios under time-sequence conditions. Moreover, the average decision-making accuracy was improved, and the average decision-making time was reduced to 405.3 ms compared with the two classical reinforcement learning algorithms.
微信 | 公众平台 
