High-precision SAR image target detection method and system based on particle filtering

Abstract

The present invention provides a high-precision SAR image target detection method and system based on particle filtering, the method comprising: using a context-aware algorithm to perform context-aware operations; feature extraction and target initialization; performing context-aware operations to obtain environmental adaptability information; continuous state estimation and tracking; comprehensive update of state estimation; using a feedback mechanism to perform feedback operations; and adaptive adjustment. The present invention solves the technical problems of poor processing speed, real-time performance, accuracy, robustness, adaptability, and flexibility, as well as high sensitivity to shielding and interference.

Description

Technical Field

[0001] The present invention relates to the field of radar image detection and recognition and artificial intelligence algorithms, and in particular to a high-precision SAR image target detection method and system based on particle filtering. Background Art

[0002] Synthetic aperture radar (SAR) technology, due to its unique all-weather and all-day working ability, plays an important role in military reconnaissance, geographic mapping, environmental monitoring and other fields. SAR image target detection, especially in complex terrain and changeable weather conditions, is crucial for understanding and analyzing radar data. For example, the existing invention patent application document "Multi-channel synthetic aperture radar moving target detection method based on subdomain adaptation" with publication number CN116413723A, the existing method includes: obtaining simulation and measured data of multi-channel synthetic aperture radar, making network training and verification data sets, adding labels to clutter and moving targets; constructing a subdomain adaptive residual network for ground moving target detection; preprocessing network training and verification data, inputting subdomain adaptive residual network, and saving the optimal network after training; sliding the window of the radar data to be tested to make a network test data set; inputting the preprocessed network test data into the saved network to obtain the predicted label; and combining all the predicted results to obtain the ground moving target detection result. However, the existing SAR target detection technology faces many challenges, such as slow processing speed and insufficient detection accuracy for small or partially obscured targets. In recent years, with the development of machine learning and deep learning technology, especially in the field of image processing, SAR target detection has brought new breakthrough opportunities. In particular, particle filtering, as a method for effectively processing nonlinear and non-Gaussian distribution data, has shown its potential in dynamic target tracking. For example, the existing invention patent application document "A SAR Image Road Extraction Method" with publication number CN102831423A, the existing method includes: extracting the characteristics of the road on the SAR image based on the line target detection operator of the mean ratio, obtaining the road line feature response map and the road line feature direction map; selecting a point on the road in the SAR image as the starting point of the particle filter; based on the starting point, using the particle filter method to track the seed point on the road; using the road line feature response map as the image energy in the Snake model, and using the seed point as the initial value of the Snake model for fitting, to obtain a road target. However, in the target detection application of SAR images, how to make full use of the advantages of particle filtering, especially in improving real-time processing capabilities and adaptability to complex environments, is still a key problem to be solved. The existence of this challenge highlights the urgent need to develop more efficient and accurate new algorithms for SAR image detection.

[0003] The target detection technology of SAR images in the prior art has the following disadvantages:

[0004] Insufficient processing speed and real-time performance: Existing SAR target detection algorithms often face the problem of slow processing speed when processing large amounts of or high-resolution images, making it difficult to achieve real-time or near-real-time target detection.

[0005] Insufficient accuracy and robustness: Existing technologies often have difficulty accurately identifying small, fast-moving targets or those in complex backgrounds, especially in changeable weather and complex terrain conditions.

[0006] Sensitivity to obstruction and interference: Current technologies are sensitive to partial obstruction or interference factors of the target (such as terrain shadows and weather effects), which can easily lead to false detection or missed detection.

[0007] Insufficient adaptability and flexibility: Existing methods lack sufficient adaptability and flexibility when processing different types of SAR images, and a large number of parameter adjustments are required for different application scenarios.

[0008] In summary, the existing technology has technical problems such as poor processing speed, real-time performance, accuracy, robustness, adaptability and flexibility, as well as high sensitivity to shielding and interference. Summary of the invention

[0009] The technical problem to be solved by the present invention is how to solve the technical problems in the prior art of poor processing speed, real-time performance, accuracy, robustness, adaptability and flexibility, and high sensitivity to shielding and interference.

[0010] The present invention adopts the following technical solutions to solve the above technical problems: A high-precision SAR image target detection method based on particle filtering includes:

[0011] S1. Collect input data, use context-aware algorithms, perform context-aware operations based on the input data, perform behavior adjustment operations, dynamic parameter adjustment, and interactive learning based on the context information of the input data, and thereby obtain enhanced information for system environment understanding;

[0012] S2. In the initial stage of target detection, the convolutional neural network (CNN) is used to extract applicable features from the SAR image of the input data. Based on the applicable features, the population-based search algorithm (PSO) is used to execute a global search strategy to estimate the initial rough position of the target and the rough position of the re-entry field of view to obtain the initial features of the target.

[0013] S3, using the PF module to set a group of particles to represent the probability distribution of the target state, and updating the particles according to the initial characteristics of the target in the SAR image to obtain the target state continuous tracking estimation data;

[0014] S4, performing context perception according to the context information of the SAR image to dynamically adjust the model parameters of the convolutional neural network CNN and the PF module;

[0015] S5. Use the initial features of the target to guide the state estimation operation of the PF module to obtain state estimation data. According to the state estimation data and model parameters, a feedback mechanism is used to perform feedback operation, match the state estimation data with context information, and adaptively adjust the model parameters of the convolutional neural network CNN and the PF module to complete the state estimation comprehensive update operation, obtain a suitable SAR image target detection model, and detect SAR image targets based on it.

[0016] The present invention adopts a fusion strategy of particle filtering (PF), convolutional neural network (CNN) and context-aware algorithm through improved algorithms and data processing technology. By fusing the three algorithms, their complementary advantages can be utilized to achieve more accurate target state estimation and reduce the limitations that may be caused by a single method. The present invention significantly improves the performance of multi-target tracking of SAR images, especially in the face of clutter and dynamically changing scenes. Combining multiple algorithms can improve the system's adaptability to noise and dynamic environmental changes and enhance the robustness of the entire tracking system.

[0017] The present invention brings out the role of each model in the field in which it is best. The target detection accuracy of the SAR image target detection system in complex environments (such as changeable weather and complex terrain) is improved, while the detection capability of small or fast-moving targets is enhanced, and the accuracy and robustness are enhanced. The present invention can quickly process a large number of or high-resolution SAR images, realize real-time or near real-time target detection, and improve processing speed and real-time performance. When processing different types of SAR images, the present invention does not require a large number of parameter adjustments for different application scenarios, thereby improving the adaptability and flexibility of the SAR image target detection system.

[0018] In a more specific technical solution, step S1 includes:

[0019] S11, performing behavior adjustment operations according to the context information of the input data, thereby obtaining environmental adaptability information;

[0020] S12, obtaining context change information in the input data, and adjusting dynamic parameters accordingly;

[0021] S13. Tune the context-aware algorithm by integrating user and system feedback to achieve interactive learning.

[0022] The present invention utilizes context-aware algorithms to consider additional information in the decision-making process, thereby enhancing the system's understanding of its operating environment. The present invention utilizes context-aware algorithms to analyze and adapt to specific environments (such as cities, forests, etc.) of SAR images, and adjusts detection strategies to reduce false detections and improve the adaptability of the system. The present invention can automatically adjust processing parameters to adapt to different types of SAR images and diverse application scenarios, reduce manual intervention, and enhance system adaptability and flexibility. At the same time, the present invention dynamically adjusts the parameters of CNN and PF according to the contextual information of the SAR image (such as environment type, weather conditions), thereby enhancing the adaptability and robustness of target detection in a changing environment.

[0023] In a more specific technical solution, in step S13, the following logic is used to perform context-aware operations to obtain system environment understanding enhancement information:

[0024] θ=f(E,θ prev )

[0025] In the formula, θ is the adjusted parameter, E is the environmental context information, and θ prev is the previous parameter value, and f is the function that adjusts the parameter according to the environment.

[0026] In a more specific technical solution, step S2 includes:

[0027] S21. Construct a convolutional neural network (CNN) to identify pattern information in SAR images, extract SAR features, and perform comprehensive classification operations on the SAR features to obtain applicable features.

[0028] S22, evaluating the current position of each particle based on applicable features, target saliency, and image similarity;

[0029] S23, updating the current speed and current position of the particle according to the optimal position information to explore the preset search space;

[0030] S24, selecting and tuning PSO parameters, the PSO parameters include: inertia weight, individual learning factor and social learning factor.

[0031] The present invention appropriately selects PSO parameters, such as inertia weight, individual learning factor and social learning factor, through parameter adjustment operation to optimize the convergence speed and search capability.

[0032] The particle filter used in the present invention has a high nonlinear state estimation capability when the target feature changes or motion has nonlinear characteristics, and the target in the SAR image is more accurately identified and located. The deep learning feature extraction of the convolutional neural network is used to automatically identify target features from complex SAR images and improve the accuracy of target detection. The present invention optimizes the algorithm to better handle partial shielding or interference factors of the target, reduce false detection and missed detection, and improve resistance to shielding and interference.

[0033] In a more specific technical solution, in step S21, the convolutional neural network CNN performs a comprehensive classification operation on the SAR features through a fully connected layer using the following logic:

[0034]

[0035] In the formula, is the eigenvalue at position (i, j) after the lth convolution layer, is the convolution kernel of layer l, in The input of the soldier (l-1) layer, B (l) is the bias term, σ is the ReLU activation function, M represents the number of rows of the image, and N represents the number of columns of the image. Corresponding to the height and width of the image, m and n: represent the position of a specific element in the M×N matrix or image, m represents the index of a specific row, and n represents the index of a specific column.

[0036] The present invention utilizes the deep learning feature extraction capability of CNN to identify potential target features from complex image data, providing an accurate starting point for the particle filter, reducing subsequent blind searches in the state space, and improving the efficiency and accuracy of detection, so that the present invention can accurately identify small, fast-moving targets or targets in complex backgrounds under changeable weather and complex terrain conditions.

[0037] In a more specific technical solution, step S22 includes:

[0038] S221. Use no less than 2 particles to represent the target potential position and target speed;

[0039] S222. Based on the saliency of the target and the image similarity, a fitness function is constructed to evaluate the current position of each particle.

[0040] In a more specific technical solution, in step S23, the optimal position information includes: an individual optimal position and a global optimal position.

[0041] In a more specific technical solution, in step S23, the inertia weight of the particle is updated using the following logic:

[0042]

[0043] In the formula, is the weight of the ith particle at time t, is the state of a given particle Observation data z t probability.

[0044] The present invention is designed to adaptively adjust the parameters of the entire fusion system, including the parameters of the CNN model, the number of particles of the PF and the resampling strategy, based on real-time performance feedback, so as to ensure that the system maintains optimal performance in the face of different scenarios and target behaviors.

[0045] In a more specific technical solution, in step S3, the particles are updated using the following logic to obtain target state continuous tracking estimation data:

[0046] S t =Combine(F CNN (I t ),PF(S t-1 ,O t ))

[0047] In the formula, S t is the system state at time t, F CNN is the feature extracted by CNN, I t is the input image at time t, PF is the particle filter, s t-1 is the state at the previous moment, O t is the observation data.

[0048] In a more specific technical solution, the high-precision SAR image target detection system based on particle filtering includes:

[0049] An input data context awareness module is used to collect input data, use context awareness algorithms, perform context awareness operations based on the input data, and perform behavior adjustment operations, dynamic parameter adjustment, and interactive learning based on the context information of the input data, thereby obtaining enhanced information for system environment understanding;

[0050] The global search module is used to extract applicable features from the SAR image of the input data using the convolutional neural network (CNN) at the initial stage of target detection. According to the applicable features, the global search strategy is executed using the population-based search algorithm (PSO) to estimate the initial rough position of the target and the rough position of the re-entry field of view to obtain the initial features of the target. The global search module is connected to the input data context perception module.

[0051] The tracking estimation module is used to set a group of particles representing the probability distribution of the target state by using the PF module, and update the particles according to the initial characteristics of the target in the SAR image to obtain the continuous tracking estimation data of the target state. The tracking estimation module is connected with the global search module;

[0052] A model parameter adjustment module is used to perform context perception according to the context information of the SAR image, so as to dynamically adjust the model parameters of the convolutional neural network CNN and the PF module. The model parameter adjustment module is connected to the global search module;

[0053] The SAR image target detection module is used to use the initial features of the target to guide the state estimation operation of the PF module to obtain state estimation data. According to the state estimation data and model parameters, the feedback mechanism is used to perform feedback operation, match the state estimation data with context information, and adaptively adjust the model parameters of the convolutional neural network CNN and the PF module to complete the state estimation comprehensive update operation, and obtain a suitable SAR image target detection model to detect SAR image targets. The SAR image target detection module is connected with the model parameter adjustment module and the global search module.

[0054] Compared with the prior art, the present invention has the following advantages:

[0055] The present invention adopts a fusion strategy of particle filtering (PF), convolutional neural network (CNN) and context-aware algorithm through improved algorithms and data processing technology. By fusing the three algorithms, their complementary advantages can be utilized to achieve more accurate target state estimation and reduce the limitations that may be caused by a single method. The present invention significantly improves the performance of multi-target tracking of SAR images, especially in the face of clutter and dynamically changing scenes. Combining multiple algorithms can improve the system's adaptability to noise and dynamic environmental changes and enhance the robustness of the entire tracking system.

[0056] The present invention brings out the role of each model in the field in which it is best. The target detection accuracy of the SAR image target detection system in complex environments (such as changeable weather and complex terrain) is improved, while the detection capability of small or fast-moving targets is enhanced, and the accuracy and robustness are enhanced. The present invention can quickly process a large number of or high-resolution SAR images, realize real-time or near real-time target detection, and improve processing speed and real-time performance. When processing different types of SAR images, the present invention does not require a large number of parameter adjustments for different application scenarios, thereby improving the adaptability and flexibility of the SAR image target detection system.

[0057] The present invention utilizes context-aware algorithms to consider additional information in the decision-making process, thereby enhancing the system's understanding of its operating environment. The present invention utilizes context-aware algorithms to analyze and adapt to specific environments (such as cities, forests, etc.) of SAR images, and adjusts detection strategies to reduce false detections and improve the adaptability of the system. The present invention can automatically adjust processing parameters to adapt to different types of SAR images and diverse application scenarios, reduce manual intervention, and enhance system adaptability and flexibility. At the same time, the present invention dynamically adjusts the parameters of CNN and PF according to the contextual information of the SAR image (such as environment type, weather conditions), thereby enhancing the adaptability and robustness of target detection in a changing environment.

[0058] The present invention appropriately selects PSO parameters, such as inertia weight, individual learning factor and social learning factor, through parameter adjustment operation to optimize the convergence speed and search capability.

[0059] The particle filter used in the present invention has a high nonlinear state estimation capability when the target feature changes or motion has nonlinear characteristics, and the target in the SAR image is more accurately identified and located. The deep learning feature extraction of the convolutional neural network is used to automatically identify target features from complex SAR images and improve the accuracy of target detection. The present invention optimizes the algorithm to better handle partial shielding or interference factors of the target, reduce false detection and missed detection, and improve resistance to shielding and interference.

[0060] The present invention utilizes the deep learning feature extraction capability of CNN to identify potential target features from complex image data, providing an accurate starting point for the particle filter, reducing subsequent blind searches in the state space, and improving the efficiency and accuracy of detection, so that the present invention can accurately identify small, fast-moving targets or targets in complex backgrounds under changeable weather and complex terrain conditions.

[0061] The present invention is designed to adaptively adjust the parameters of the entire fusion system, including the parameters of the CNN model, the number of particles of the PF and the resampling strategy, based on real-time performance feedback, so as to ensure that the system maintains optimal performance in the face of different scenarios and target behaviors.

[0062] The present invention solves the technical problems existing in the prior art of poor processing speed, real-time performance, accuracy, robustness, adaptability and flexibility, and high sensitivity to shielding and interference. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a schematic diagram of the basic steps of a high-precision SAR image target detection method based on particle filtering according to Embodiment 1 of the present invention;

[0064] Figure 2This is a schematic diagram of data flow processing of a high-precision SAR image target detection method based on particle filtering according to Example 1 of the present invention. DETAILED DESCRIPTION

[0065] 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 clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are 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.

[0066] Example 1

[0067] like Figure 1 and Figure 2 As shown, the high-precision SAR image target detection method based on particle filtering provided by the present invention includes the following basic steps:

[0068] S1. Use context-aware algorithms to perform context-aware operations;

[0069] The context-aware algorithm used in this embodiment is an intelligent processing technology that considers additional information in the decision-making process to enhance the system's understanding of its operating environment.

[0070] In this embodiment, environmental adaptability: the algorithm adjusts its behavior according to the contextual information of the input data (such as environment type, time information); dynamic parameter adjustment: automatically adjusts processing parameters according to context changes, such as adjusting the sensitivity of the detection algorithm or the strategy of feature extraction; interactive learning: integrates user or system feedback to continuously optimize and adjust the performance of the algorithm.

[0071] θ=f(E,θ prev )

[0072] Among them, θ is the adjusted parameter, E is the environmental context information, and θ prev is the previous parameter value, and f is the function that adjusts the parameter according to the environment.

[0073] S2, feature extraction and target initialization;

[0074] In the initial stage of target detection in this embodiment, CNN is used to extract key features from SAR images. CNN's deep learning feature extraction capability is particularly good at identifying potential target features from complex image data. This step is important because it provides an accurate starting point for the particle filter, reduces subsequent blind searches in the state space, and improves the efficiency and accuracy of detection;

[0075] In this embodiment, PSO is a population-based search algorithm inspired by the foraging behavior of bird flocks. In the field of SAR image tracking, PSO is used as a global search strategy to estimate the rough position of the target in the image when it initially or re-enters the field of view; in this embodiment, during the initialization process, a certain number of particles are randomly distributed in the search space, and each particle represents the potential position and velocity of the target; in the process of defining the fitness function of this embodiment, a fitness function based on the saliency or image similarity measure of the target is defined to evaluate the position of each particle; in the speed and position update operation of this embodiment, the particles update their speed and position according to the individual best position and the global best position to explore the search space; in the parameter adjustment operation of this embodiment, PSO parameters such as inertia weight, individual learning factor and social learning factor are appropriately selected to optimize the convergence speed and search ability;

[0076] In this embodiment, the particle weight update formula is as follows:

[0077]

[0078] in, is the weight of the ith particle at time t, is the state of a given particle Observation data z t probability;

[0079] In this embodiment, a convolutional neural network (CNN) is constructed. The convolutional neural network (CNN) is a deep learning architecture widely used in the field of image processing. The convolutional neural network effectively recognizes patterns and features in images by simulating the processing method of the human visual system.

[0080] Convolutional neural network CNN includes but is not limited to: convolution layer, activation function, pooling layer, fully connected layer; in the convolution layer of this embodiment, a series of learnable filters are used to automatically extract features in the image. Each filter is responsible for capturing local features of the image, such as edges or textures; the activation function ReLU (Rectified Linear Unit) is used to increase the expressive power of the model; the pooling layer in this embodiment is used to reduce the amount of calculation by reducing the spatial size of the feature, and to a certain extent improve the translation invariance of the feature; in this embodiment, the fully connected layer is used after the convolution and pooling layers, and the network integrates and classifies the extracted features through the fully connected layer:

[0081]

[0082] in, is the eigenvalue at position (i, j) after the lth convolution layer, is the convolution kernel of layer l, in The input of the soldier (l-1) layer, B (l) is the bias term, σ is the ReLU activation function;

[0083] S3, continuous state estimation and tracking;

[0084] In this embodiment, once the initial features of the target are extracted by CNN, PF takes over the subsequent detection process. PF uses a set of particles to represent the probability distribution of the target state, each particle represents a possible state hypothesis. As new SAR image frames arrive, PF updates these particles based on the observed data, providing a continuous estimate of the target state:

[0085] S t =Combine(F CNN (I t ),PF(S t-1 ,O t ))

[0086] Among them, S t is the system state at time t, F CNN is the feature extracted by CNN, I t is the input image at time t, PF is the particle filter, S t-1 is the state at the previous moment, O t is the observational data;

[0087] S4, perform context perception and obtain environmental adaptability information;

[0088] In this embodiment, the context-aware algorithm plays a role in the entire detection process, dynamically adjusting the parameters of CNN and PF according to the context information of the SAR image (such as environment type, weather conditions). This method enhances the adaptability and robustness of target detection in a changing environment.

[0089] S5, comprehensive update of state estimation;

[0090] In each time step of this embodiment, the feature information provided by CNN, the dynamic state estimation of PF and the adaptability analysis of the context-aware algorithm are used in combination; in this embodiment, the feature information provided by CNN is used to guide the state estimation of PF, and the output of the context-aware algorithm is used to adjust the parameters of CNN and PF, thereby completing the comprehensive update of the state estimation.

[0091] S6. Use feedback mechanism to perform feedback operation;

[0092] In this embodiment, a feedback mechanism is introduced to ensure the robustness of the system; in this embodiment, if the state estimate provided by the PF does not match the expected target dynamics or context information, the system can adjust the parameters of the CNN and PF, or even trigger re-extraction of features;

[0093] S7, adaptive adjustment;

[0094] In this embodiment, the parameters of the entire fusion system, including but not limited to: the parameters of the CNN model, the number of particles of the PF, and the resampling strategy, are designed to be adaptively adjusted according to real-time performance feedback. This ensures that the system maintains optimal performance when facing different scenarios and target behaviors.

[0095] In summary, the present invention adopts a fusion strategy of particle filtering (PF) with convolutional neural network (CNN) and context-aware algorithm through improved algorithms and data processing technology. By fusing the three algorithms, their complementary advantages can be utilized to achieve more accurate target state estimation and reduce the limitations that may be caused by a single method. The present invention significantly improves the performance of multi-target tracking of SAR images, especially in the face of clutter and dynamically changing scenes. Combining multiple algorithms can improve the system's adaptability to noise and dynamic environmental changes and enhance the robustness of the entire tracking system.

[0096] The present invention brings out the role of each model in the field in which it is best. The target detection accuracy of the SAR image target detection system in complex environments (such as changeable weather and complex terrain) is improved, while the detection capability of small or fast-moving targets is enhanced, and the accuracy and robustness are enhanced. The present invention can quickly process a large number of or high-resolution SAR images, realize real-time or near real-time target detection, and improve processing speed and real-time performance. When processing different types of SAR images, the present invention does not require a large number of parameter adjustments for different application scenarios, thereby improving the adaptability and flexibility of the SAR image target detection system.

[0097] The present invention utilizes context-aware algorithms to consider additional information in the decision-making process, thereby enhancing the system's understanding of its operating environment. The present invention utilizes context-aware algorithms to analyze and adapt to specific environments (such as cities, forests, etc.) of SAR images, and adjusts detection strategies to reduce false detections and improve the adaptability of the system. The present invention can automatically adjust processing parameters to adapt to different types of SAR images and diverse application scenarios, reduce manual intervention, and enhance system adaptability and flexibility. At the same time, the present invention dynamically adjusts the parameters of CNN and PF according to the contextual information of the SAR image (such as environment type, weather conditions), thereby enhancing the adaptability and robustness of target detection in a changing environment.

[0098] The present invention appropriately selects PSO parameters, such as inertia weight, individual learning factor and social learning factor, through parameter adjustment operation to optimize the convergence speed and search capability.

[0099] The particle filter used in the present invention has a high nonlinear state estimation capability when the target feature changes or motion has nonlinear characteristics, and the target in the SAR image is more accurately identified and located. The deep learning feature extraction of the convolutional neural network is used to automatically identify target features from complex SAR images and improve the accuracy of target detection. The present invention optimizes the algorithm to better handle partial shielding or interference factors of the target, reduce false detection and missed detection, and improve resistance to shielding and interference.

[0100] The present invention utilizes the deep learning feature extraction capability of CNN to identify potential target features from complex image data, providing an accurate starting point for the particle filter, reducing subsequent blind searches in the state space, and improving the efficiency and accuracy of detection, so that the present invention can accurately identify small, fast-moving targets or targets in complex backgrounds under changeable weather and complex terrain conditions.

[0101] The present invention is designed to adaptively adjust the parameters of the entire fusion system, including the parameters of the CNN model, the number of particles of the PF and the resampling strategy, based on real-time performance feedback, so as to ensure that the system maintains optimal performance in the face of different scenarios and target behaviors.

[0102] The present invention solves the technical problems existing in the prior art of poor processing speed, real-time performance, accuracy, robustness, adaptability and flexibility, and high sensitivity to shielding and interference.

[0103] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A high-precision SAR image target detection method based on particle filtering, characterized in that: The method comprises: S1. Collect input data, use context-aware algorithms, perform context-aware operations based on the input data, perform behavior adjustment operations, dynamic parameter adjustment, and interactive learning based on context information of the input data, and thereby obtain system environment understanding enhancement information; Wherein, the step S1 comprises: S11, performing the behavior adjustment operation according to the context information of the input data, thereby obtaining environmental adaptability information; S12, obtaining context change information in the input data, and adjusting dynamic parameters accordingly; S13, tuning the context-aware algorithm by integrating user and system feedback to complete the interactive learning; In step S13, the context awareness operation is performed using the following logic to obtain the system environment understanding enhancement information: θ=f(E,θ prev ) In the formula, θ is the adjusted parameter, E is the environmental context information, and θ prev is the previous parameter value, and f is the function that adjusts the parameter according to the environment; S2. In the initial stage of target detection, a convolutional neural network (CNN) is used to extract applicable features from the SAR image of the input data. Based on the applicable features, a population-based search algorithm (PSO) is used to execute a global search strategy to estimate the initial rough position of the target and the rough position of the re-entry field of view to obtain the initial features of the target. The step S2 comprises: S21, constructing the convolutional neural network CNN, identifying pattern information in the SAR image, extracting SAR features, and performing a comprehensive classification operation on the SAR features to obtain the applicable features; S22, evaluating the current position of each particle according to the applicable features, the saliency of the target, and the image similarity; The step S22 comprises: S221, using at least two of the particles to represent the target potential position and target speed; S222, constructing a fitness function based on the saliency and image similarity of the target, and evaluating the current position of each particle; S23, updating the current speed and the current position of the particle according to the optimal position information to explore a preset search space; Wherein, in step S23, the inertia weight of the particle is updated using the following logic: In the formula, is the weight of the ith particle at time t, is the state of a given particle Observation data z t probability; S24, selecting and optimizing PSO parameters, wherein the PSO parameters include: inertia weight, individual learning factor, and social learning factor; S3, using the PF module to set a group of particles to represent the probability distribution of the target state, and updating the particles according to the initial characteristics of the target in the SAR image to obtain the target state continuous tracking estimation data; S4, performing context perception according to the context information of the SAR image to dynamically adjust the model parameters of the convolutional neural network CNN and the PF module; S5. Use the initial features of the target to guide the state estimation operation of the PF module to obtain state estimation data. According to the state estimation data and the model parameters, use the feedback mechanism to perform feedback operation, match the state estimation data with context information, and adaptively adjust the model parameters of the convolutional neural network CNN and the PF module to complete the state estimation comprehensive update operation, obtain a suitable SAR image target detection model, and detect SAR image targets based on it.

2. The high-precision SAR image target detection method based on particle filtering according to claim 1 is characterized in that: In step S21, the convolutional neural network CNN performs the comprehensive classification operation on the SAR features through a fully connected layer using the following logic: In the formula, is the eigenvalue at position (i, j) after the lth convolution layer, is the convolution kernel of layer l, in The input of the soldier (l-1) layer, B (l) is the bias term, σ is the ReLU activation function, M represents the number of rows of the image, N represents the number of columns of the image, m represents the index of a specific row, and n represents the index of a specific column.

3. The high-precision SAR image target detection method based on particle filtering according to claim 1 is characterized in that: In the step S23, the optimal position information includes: an individual optimal position and a global optimal position.

4. The high-precision SAR image target detection method based on particle filtering according to claim 1 is characterized in that: In step S3, the particles are updated using the following logic to obtain the target state continuous tracking estimation data: S t =Combine(F CNN (I t ),PF(S t-1 ,O t )) In the formula, S t is the system state at time t, F CNN is the feature extracted by CNN, I t is the input image at time t, PF is the particle filter, S t-1 is the state at the previous moment, O t is the observation data.

5. A high-precision SAR image target detection system based on particle filtering, used to execute the high-precision SAR image target detection method based on particle filtering according to any one of claims 1 to 4, characterized in that: The system comprises: An input data context perception module, for collecting input data, using a context perception algorithm, performing context perception operations based on the input data, so as to perform behavior adjustment operations, dynamic parameter adjustment, and interactive learning based on the context information of the input data, thereby obtaining system environment understanding enhancement information; A global search module, used for extracting applicable features from the SAR image of the input data using a convolutional neural network (CNN) at the initial stage of target detection, and executing a global search strategy using a population-based search algorithm (PSO) according to the applicable features to estimate the initial rough position of the target and the rough position of re-entering the field of view to obtain the initial features of the target, wherein the global search module is connected to the input data context perception module; A tracking estimation module, used to set a group of particles representing the probability distribution of the target state by using the PF module, and update the particles according to the initial characteristics of the target in the SAR image to obtain continuous tracking estimation data of the target state, wherein the tracking estimation module is connected to the global search module; A model parameter adjustment module, used for performing context perception according to the context information of the SAR image, so as to dynamically adjust the model parameters of the convolutional neural network CNN and the PF module, wherein the model parameter adjustment module is connected to the global search module; The SAR image target detection module is used to guide the state estimation operation of the PF module by using the initial features of the target to obtain state estimation data, and to perform feedback operation by using a feedback mechanism according to the state estimation data and the model parameters, so as to match the state estimation data with context information, and to adaptively adjust the model parameters of the convolutional neural network CNN and the PF module, thereby completing the state estimation comprehensive update operation, and obtaining a suitable SAR image target detection model, so as to detect SAR image targets. The SAR image target detection module is connected to the model parameter adjustment module and the global search module.

Images