Image processing analysis method for underwater radar signal feedback information

Abstract

The invention discloses an image processing analysis method for underwater radar signal feedback information. The method comprises the following steps: setting a multi-source signal acquisition module for synchronously acquiring original feedback signal information received after a plurality of underwater radar transmitting signals; the preprocessed underwater radar signals are converted into corresponding image representation forms through an image generation module; an image fusion and matching module is set, and fusion processing is carried out on multiple groups of underwater radar detection images; an image analysis module is set, and comprehensive analysis processing is carried out on the fused image containing the same object and seabed contour information; a target detection and identification module is arranged to carry out detection and identification analysis on the same object, and dangerous sources possibly influencing ship navigation are identified; establishing a user interaction interface module with a labeling function, wherein the user interaction interface module is used for labeling and displaying the image content on an interface; and a result output module is established, and the result obtained by the target detection and identification module is sorted and output.

Description

Technical Field

[0001] The invention belongs to the technical field of underwater radar detection, and specifically refers to an image processing and analysis method for underwater radar signal feedback information. Background Art

[0002] In the field of underwater detection and monitoring, underwater radar is an important detection method that can obtain feedback image information of underwater targets. However, with the increasing complexity and diversification of underwater detection tasks, higher requirements are placed on the analysis and processing of underwater radar images.

[0003] Traditional underwater radar image analysis mainly relies on manual interpretation. Operators use their experience and expertise to identify and judge the target features in the image. This manual analysis method has many limitations. First, manual analysis is inefficient. When faced with a large amount of image data, it is difficult to complete the processing in a short time, and it cannot meet underwater monitoring tasks with high real-time requirements, such as real-time target warning in underwater military defense and continuous regional exploration in marine resource exploration. Secondly, the accuracy of manual analysis is easily affected by subjective factors of operators. Differences in experience, fatigue, and concentration of different personnel will lead to deviations in the analysis results of the same image, thereby affecting the correct judgment of underwater targets and subsequent decision-making.

[0004] With the development of computer technology and image processing technology, some semi-automatic image analysis methods have begun to be applied. For example, the method based on simple feature extraction and threshold setting extracts features such as grayscale value and texture in the image and compares them with the preset threshold to try to distinguish the target from the background or identify a specific type of target. However, this method has poor adaptability. For complex and changeable underwater environments and target objects of different shapes, the preset feature threshold is difficult to be universally applied. Light changes in the underwater environment, differences in water turbidity, different postures and occlusion of the target, etc. will make the method based on fixed thresholds invalid, resulting in a large number of misjudgments and missed judgments.

[0005] At the same time, in terms of result display, traditional analysis methods can usually only provide simple text reports or basic image annotations, lacking intuitive, detailed and easy-to-understand visual display methods. This requires operators to spend extra time and energy to interpret and understand the analysis results after obtaining them, which is not conducive to rapid response and decision-making.

[0006] In summary, the existing underwater radar image analysis technology has obvious deficiencies in analysis efficiency, accuracy, target classification precision and result display effect. There is an urgent need for an innovative technology that can automatically, efficiently and accurately analyze the object types in underwater radar images and display them on the screen in an intuitive and effective way to meet the growing needs of underwater detection and monitoring. Summary of the invention

[0007] In order to overcome the problems of the prior art, the present invention designs an image processing and analysis method for underwater radar signal feedback information, which can utilize the underwater radar signal feedback information, through image processing and analysis, to achieve target detection and identification, seabed contour drawing, dynamic analysis and early warning, and data interaction, thereby improving the multi-faceted efficiency of underwater operations.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows: an image processing and analysis method for underwater radar signal feedback information, comprising the following steps:

[0009] Step S1: Setting a multi-source signal acquisition module to synchronously acquire the original feedback signal information received after multiple underwater radars transmit signals, and converting them into digital signal formats for storage;

[0010] Step S2: Establish a signal preprocessing module to perform preliminary processing on the digital signals corresponding to each underwater radar;

[0011] Step S3: converting the pre-processed underwater radar signals into corresponding image representation forms through an image generation module to obtain multiple groups of underwater radar detection images;

[0012] Step S4: Setting up an image fusion and matching module to fuse multiple sets of underwater radar detection images and identify the same object in the image, while extracting information related to the seabed contour;

[0013] Step S5: Setting up an image analysis module to perform comprehensive analysis and processing on the fused images containing the same object and the seabed contour information;

[0014] Step S6: Setting up a target detection and recognition module, based on the processing results of the image analysis module, performing detailed detection and recognition analysis on the same object, and identifying the danger sources that may affect the navigation of the ship; evaluating the recognition accuracy during the detection and recognition process, and making inferences and giving corresponding probabilities when the object type cannot be accurately determined, and updating the probability with continuous detection;

[0015] Step S7: Establish a user interaction interface module with a labeling function, which is used to label and display the image content on the interface, and visually label and display the results obtained by the target detection and recognition module, including object information, seabed contour information, hazard source information, recognition accuracy, and inference probability information;

[0016] Step S8: Establish a result output module to organize and output the results obtained by the target detection and recognition module, present them to the user in a visual manner, and support data storage and export functions.

[0017] Furthermore, in step S2, when the signal preprocessing module performs preliminary processing on the digital signals corresponding to each underwater radar, the following sub-steps are included:

[0018] Step S2-1: Use an adaptive bandpass filtering algorithm to remove noise components outside a specific frequency band in each digital signal, dynamically set the filtering frequency band range according to the operating frequency and signal characteristics of each underwater radar, and accurately suppress various types of noise interference;

[0019] Step S2-2: normalize the denoised digital signals and adjust the signal amplitude according to a unified standard scale to ensure that the images subsequently generated by different underwater radar signals are comparable in terms of intensity, etc., to facilitate subsequent fusion and analysis operations.

[0020] Furthermore, in step S4, when the image fusion and matching module fuses multiple groups of underwater radar detection images and identifies the same object corresponding to the image and extracts information related to the seabed contour, the following sub-steps are included:

[0021] Step S4-1: spatially register multiple groups of underwater radar detection images. Based on the installation position, attitude information and known underwater geographic coordinates of each underwater radar, objects and seabed contours in different radar images are mapped to the same spatial coordinate system through geometric transformation operations such as coordinate transformation, rotation and translation, laying the foundation for subsequent recognition of the same object and extraction of seabed contours.

[0022] Step S4-2: Using a feature extraction algorithm to extract features from each registered image, the extracted features include local features of the image, namely, scale-invariant feature transform (SIFT) features, speeded-up robust features (SURF) features, and global features, including color histograms and texture features, to obtain rich feature description information of each image;

[0023] Step S4-3: Using a feature matching algorithm, based on the extracted features, find the matching relationship of feature points corresponding to the same object between different images, and by calculating the Euclidean distance between feature vectors as a distance metric, screen out matching pairs that meet the similarity threshold, and determine the areas belonging to the same object in different images based on the matching pairs;

[0024] Step S4-4: performing fusion processing on the successfully matched regions of the same object, comprehensively considering the information of the object at different viewing angles and different resolutions in each image, and generating a fused object image representation through weighted averaging and multi-resolution fusion methods, so that it contains more comprehensive and accurate object feature information;

[0025] Step S4-5: For the seabed contour part, the edge detection algorithm is combined with the seabed terrain feature model to extract the seabed contour line from the registered image. By fusing and optimizing the seabed contour information in multiple groups of images, high-precision seabed contour map information is constructed.

[0026] Furthermore, in step S5, when the image analysis module performs comprehensive analysis on the fused images containing the same object and the seabed contour information, the following sub-steps are included:

[0027] Step S5-1: for each frame of fused image, use an image segmentation algorithm to divide the image into different regions, and use a threshold segmentation, region growing or cluster analysis method to preliminarily distinguish possible target regions and background regions;

[0028] Step S5-2: extract multi-dimensional features from each region, including grayscale features, i.e., mean, variance, and histogram distribution. Texture features are obtained through grayscale co-occurrence matrix and local binary pattern method. Shape features are obtained based on contour analysis and geometric moment calculation to form feature vectors corresponding to each region of each frame image.

[0029] Step S5-3: Considering the continuous multi-frame fusion images, for the objects in the same position or related area in the image, analyze the feature changes between different frames, calculate the feature difference, change and other indicators, including the grayscale change of the target area between adjacent frames, the change trend of texture complexity, and the deformation degree of the shape, and integrate these feature information reflecting the dynamic changes of the object into the feature vector, and expand it into a comprehensive feature vector that integrates dynamic features;

[0030] Step S5-4: Correlate and analyze the seabed contour information with the object feature information to determine the relative position relationship between the object and the seabed and the movement trend of the object on the seabed topography, so as to further enrich the information dimension of the comprehensive feature vector.

[0031] Furthermore, in step S6, the target detection and recognition module performs detailed detection and recognition analysis on the same object and identifies the hazard source, and evaluates the recognition accuracy, infers the object type probability and updates it, including the following sub-steps:

[0032] Step S6-1: Construct a comprehensive training data set, collect multiple sets of continuous image data generated by historical collaborative detection of multiple underwater radars, and manually accurately annotate the same object and the corresponding seabed contour information and possible danger sources. The annotation content includes the category of the object, the accurate position, size, shape characteristics and motion state corresponding to different times in the fused image, and other multi-dimensional information. At the same time, the association information between the seabed contour characteristics and the object, as well as the type and degree of danger of the danger source are annotated, so as to construct a training data set covering multi-radar image fusion, object dynamic characteristics, seabed contour association and danger source annotation;

[0033] Step S6-2: Build a target detection and recognition model based on a deep learning framework, select a network architecture that has the ability to process multi-source data fusion, object dynamic change analysis, seabed contour association, and hazard source identification, such as a multimodal fusion convolutional neural network combined with a recurrent neural network. Use the constructed comprehensive training data set to train the model so that the model learns the mapping relationship between the multi-dimensional features of the same object in the fused image, seabed contour information, hazard source features, and various attributes of the actual target object. Use the stochastic gradient descent optimization algorithm to reasonably adjust the model parameters to improve the accuracy and robustness of the model in object detection and identification and hazard source judgment in multiple radar images and object dynamic scenes;

[0034] Step S6-3: Input the continuous multi-frame fusion image to be analyzed into the trained target detection and recognition model. The model comprehensively considers the feature information of each frame of the image and the object between frames, the seabed contour information, and outputs the accurate category, precise location coordinates, detailed size information, shape parameters and real-time motion status of the object such as speed, direction, acceleration, and at the same time identifies possible sources of danger, including the category, location, and danger range of the source of danger; in the output result, the model calculates the recognition accuracy according to the internal confidence evaluation mechanism, and when the recognition result is uncertain, based on the similarity analysis of the object features and the features of each category in the training data, the possible object type is inferred and the corresponding probability value is given; as new image frames are continuously input for detection, the model continuously updates the inferred probability of the object type based on the new feature information; reliable target detection results are screened out according to the preset confidence threshold, so as to realize the accurate analysis of the same object in multiple radar images and the effective identification of the source of danger.

[0035] Furthermore, in step S7, when the user interaction interface module with annotation function annotates and displays the image content, the following operations are included:

[0036] Use different colors, shapes or icons to distinguish and mark the identified objects, seabed contours and danger sources on the interface. Use blue lines to outline the seabed contours, green boxes to mark normal objects, and red warning icons to mark danger sources.

[0037] Provide a detailed display window for the marked information. When the user hovers the mouse over or clicks the marked area, detailed information about the object or hazard source corresponding to the area is displayed, such as category, location, size, motion status, degree of danger, recognition accuracy, and inferred probability.

[0038] It supports users to manually add, modify or delete annotation information on the interface, and associates user operation records with corresponding image data for storage for subsequent analysis and tracing.

[0039] Furthermore, in step S8, when the result output module organizes and outputs the results obtained by the target detection and recognition module, the following requirements must be met:

[0040] The report generated by the result output module includes the detailed category of the identified object, the accurate position coordinates in the underwater environment, the three-dimensional size, the shape feature description, the real-time motion status including speed, direction, acceleration, etc., and the motion trend prediction based on the historical trajectory, etc., and also includes the seabed contour map information and the relative position relationship information between the object and the seabed contour, as well as the detailed information of the hazard source, such as the category, location, hazard range, hazard degree, etc., and also includes the recognition accuracy and inferred probability information;

[0041] The charts drawn include a three-dimensional trajectory diagram of the object's position changing over time, a schematic diagram of the shape change under different viewing angles, a curve of the change of motion state parameters, a seabed contour map display, a schematic diagram of the distribution of hazard sources, etc.

[0042] The report also records the performance evaluation indicators of the model, such as target detection accuracy, recall rate, F1 value, position estimation error, motion state prediction error, and hazard source identification accuracy under fusion images, so as to fully understand the analysis effect and model performance.

[0043] Supports storage and export of result data in common formats (such as PDF, CSV, etc.) to facilitate data sharing and interaction with other systems or devices.

[0044] Furthermore, when the image fusion and matching module performs feature matching, if the feature is blurred, partially occluded, or the like, resulting in matching difficulties, a feature matching enhancement method based on deep learning is adopted to learn the semantic information of the features through a pre-trained feature matching network, thereby improving the accuracy and robustness of feature matching, and thereby more accurately identifying the same object in multiple groups of images.

[0045] Furthermore, when analyzing and processing continuous multi-frame fused images, the image analysis module adopts motion estimation technology based on the optical flow method to calculate the motion vector of the object pixel points in the image, assist in determining the object's movement direction and speed, further enrich the object's dynamic feature information, and integrate it into the comprehensive feature vector, thereby enhancing the model's accuracy in identifying the object's motion.

[0046] Furthermore, after the model training is completed, the target detection and recognition module regularly uses the newly acquired image data of collaborative detection by multiple underwater radars to perform incremental learning and optimization on the model. By analyzing the changing trends of the characteristics of the same object in the new data and the characteristics of newly emerging complex scenes, the model structure or parameters are adjusted in a targeted manner, so that the model can continuously adapt to the dynamic changes of the underwater multi-radar detection environment, ensuring that it always has good detection and recognition capabilities for different objects in multi-radar images and accurate identification capabilities for dangerous sources.

[0047] The beneficial effects of the image processing and analysis method for underwater radar signal feedback information of the present invention are as follows: through multi-source signal acquisition and processing, the target detection and recognition accuracy is improved, and the risk of misjudgment and missed judgment is reduced; accurate seabed contour drawing assists marine geology, resource exploration and engineering construction; dynamic target analysis and trajectory prediction serve multiple fields such as vehicle navigation and biological research; intelligent identification and early warning of dangerous sources effectively ensure ship navigation and marine military safety; the user interaction interface is convenient and intuitive, which is convenient for operational decision-making and information traceability management; the result output module promotes data integration and sharing, and promotes multi-faceted coordinated development of the ocean. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic flow chart of an image processing and analysis method for underwater radar signal feedback information according to the present invention;

[0049] Figure 2 A specific step diagram of step S2 of an image processing and analysis method for underwater radar signal feedback information of the present invention;

[0050] Figure 3 A specific step diagram of step S4 of an image processing and analysis method for underwater radar signal feedback information of the present invention;

[0051] Figure 4 A specific step diagram of step S5 of an image processing and analysis method for underwater radar signal feedback information of the present invention;

[0052] Figure 5 This is a specific step diagram of step S6 of an image processing and analysis method for underwater radar signal feedback information of the present invention. DETAILED DESCRIPTION

[0053] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] like Figure 1-5 As shown, the present invention is an image processing and analysis method for underwater radar signal feedback information, comprising the following steps:

[0055] Step S1: Setting a multi-source signal acquisition module to synchronously acquire the original feedback signal information received after multiple underwater radars transmit signals, and converting them into digital signal formats for storage;

[0056] Step S2: Establish a signal preprocessing module to perform preliminary processing on the digital signals corresponding to each underwater radar;

[0057] Step S3: converting the pre-processed underwater radar signals into corresponding image representation forms through an image generation module to obtain multiple groups of underwater radar detection images;

[0058] Step S4: Setting up an image fusion and matching module to fuse multiple sets of underwater radar detection images and identify the same object in the image, while extracting information related to the seabed contour;

[0059] Step S5: Setting up an image analysis module to perform comprehensive analysis and processing on the fused images containing the same object and the seabed contour information;

[0060] Step S6: Setting up a target detection and recognition module, based on the processing results of the image analysis module, performing detailed detection and recognition analysis on the same object, and identifying the danger sources that may affect the navigation of the ship; evaluating the recognition accuracy during the detection and recognition process, and making inferences and giving corresponding probabilities when the object type cannot be accurately determined, and updating the probability with continuous detection;

[0061] Step S7: Establish a user interaction interface module with a labeling function, which is used to label and display the image content on the interface, and visually label and display the results obtained by the target detection and recognition module, including object information, seabed contour information, hazard source information, recognition accuracy, and inference probability information;

[0062] Step S8: Establish a result output module to organize and output the results obtained by the target detection and recognition module, present them to the user in a visual manner, and support data storage and export functions.

[0063] Furthermore, in step S2, when the signal preprocessing module performs preliminary processing on the digital signals corresponding to each underwater radar, the following sub-steps are included:

[0064] Step S2-1: Use an adaptive bandpass filtering algorithm to remove noise components outside a specific frequency band in each digital signal, dynamically set the filtering frequency band range according to the operating frequency and signal characteristics of each underwater radar, and accurately suppress various types of noise interference;

[0065] Step S2-2: normalize the denoised digital signals and adjust the signal amplitude according to a unified standard scale to ensure that the images subsequently generated by different underwater radar signals are comparable in terms of intensity, etc., to facilitate subsequent fusion and analysis operations.

[0066] Furthermore, in step S4, when the image fusion and matching module fuses multiple groups of underwater radar detection images and identifies the same object corresponding to the image and extracts information related to the seabed contour, the following sub-steps are included:

[0067] Step S4-1: spatially register multiple groups of underwater radar detection images. Based on the installation position, attitude information and known underwater geographic coordinates of each underwater radar, objects and seabed contours in different radar images are mapped to the same spatial coordinate system through geometric transformation operations such as coordinate transformation, rotation and translation, laying the foundation for subsequent recognition of the same object and extraction of seabed contours.

[0068] Step S4-2: Using a feature extraction algorithm to extract features from each registered image, the extracted features include local features of the image, i.e., scale-invariant feature transform (SIFT) features, speeded-up robust features (SURF) features, and global features such as color histogram and texture features, to obtain rich feature description information of each image;

[0069] Step S4-3: Using a feature matching algorithm, based on the extracted features, find the matching relationship of feature points corresponding to the same object between different images, and by calculating the Euclidean distance between feature vectors as a distance metric, screen out matching pairs that meet the similarity threshold, and determine the areas belonging to the same object in different images based on the matching pairs;

[0070] Step S4-4: performing fusion processing on the successfully matched regions of the same object, comprehensively considering the information of the object at different viewing angles and different resolutions in each image, and generating a fused object image representation through weighted averaging and multi-resolution fusion methods, so that it contains more comprehensive and accurate object feature information;

[0071] Step S4-5: For the seabed contour part, the edge detection algorithm is combined with the seabed terrain feature model to extract the seabed contour line from the registered image. By fusing and optimizing the seabed contour information in multiple groups of images, high-precision seabed contour map information is constructed.

[0072] Furthermore, in step S5, when the image analysis module performs comprehensive analysis on the fused images containing the same object and the seabed contour information, the following sub-steps are included:

[0073] Step S5-1: for each frame of fused image, use an image segmentation algorithm to divide the image into different regions, and use a threshold segmentation, region growing or cluster analysis method to preliminarily distinguish possible target regions and background regions;

[0074] Step S5-2: extract multi-dimensional features from each region, including grayscale features, i.e., mean, variance, and histogram distribution. Texture features are obtained through grayscale co-occurrence matrix and local binary pattern method. Shape features are obtained based on contour analysis and geometric moment calculation to form feature vectors corresponding to each region of each frame image.

[0075] Step S5-3: Considering the continuous multi-frame fusion image, for the objects in the same position or related area in the image, analyze the feature changes between different frames, calculate the feature difference, change rate and other indicators, such as the grayscale change of the target area between adjacent frames, the change trend of texture complexity, and the deformation degree of the shape, and integrate these feature information reflecting the dynamic changes of the object into the feature vector, and expand it into a comprehensive feature vector that integrates dynamic features;

[0076] Step S5-4: Correlate and analyze the seabed contour information with the object feature information to determine the relative position relationship between the object and the seabed and the movement trend of the object on the seabed topography, so as to further enrich the information dimension of the comprehensive feature vector.

[0077] Furthermore, in step S6, the target detection and recognition module performs detailed detection and recognition analysis on the same object and identifies the hazard source, and evaluates the recognition accuracy, infers the object type probability and updates it, including the following sub-steps:

[0078] Step S6-1: Construct a comprehensive training data set, collect multiple sets of continuous image data generated by historical collaborative detection of multiple underwater radars, and manually accurately annotate the same object and the corresponding seabed contour information and possible danger sources. The annotation content includes the category of the object, the accurate position, size, shape characteristics and motion state corresponding to different times in the fused image, and other multi-dimensional information. At the same time, the association information between the seabed contour characteristics and the object, as well as the type and degree of danger of the danger source are annotated, so as to construct a training data set covering multi-radar image fusion, object dynamic characteristics, seabed contour association and danger source annotation;

[0079] Step S6-2: Build a target detection and recognition model based on a deep learning framework, select a network architecture that has the ability to process multi-source data fusion, object dynamic change analysis, seabed contour association, and hazard source identification, such as a multimodal fusion convolutional neural network combined with a recurrent neural network. Use the constructed comprehensive training data set to train the model so that the model learns the mapping relationship between the multi-dimensional features of the same object in the fused image, seabed contour information, hazard source features, and various attributes of the actual target object. Use the stochastic gradient descent optimization algorithm to reasonably adjust the model parameters to improve the accuracy and robustness of the model in object detection and identification and hazard source judgment in multiple radar images and object dynamic scenes;

[0080] Step S6-3: Input the continuous multi-frame fusion image to be analyzed into the trained target detection and recognition model. The model comprehensively considers the feature information of each frame of the image and the object between frames, the seabed contour information, and outputs the accurate category, precise location coordinates, detailed size information, shape parameters and real-time motion status of the object such as speed, direction, acceleration, and at the same time identifies possible sources of danger, including the category, location, and danger range of the source of danger; in the output result, the model calculates the recognition accuracy according to the internal confidence evaluation mechanism, and when the recognition result is uncertain, based on the similarity analysis of the object features and the features of each category in the training data, the possible object type is inferred and the corresponding probability value is given; as new image frames are continuously input for detection, the model continuously updates the inferred probability of the object type based on the new feature information; reliable target detection results are screened out according to the preset confidence threshold, so as to realize the accurate analysis of the same object in multiple radar images and the effective identification of the source of danger.

[0081] Furthermore, in step S7, when the user interaction interface module with annotation function annotates and displays the image content, the following operations are included:

[0082] Use different colors, shapes or icons to distinguish and mark the identified objects, seabed contours and danger sources on the interface. Use blue lines to outline the seabed contours, green boxes to mark normal objects, and red warning icons to mark danger sources.

[0083] Provide a detailed display window for the marked information. When the user hovers the mouse over or clicks the marked area, detailed information about the object or hazard source corresponding to the area is displayed, such as category, location, size, motion status, degree of danger, recognition accuracy, inferred probability, etc.

[0084] It supports users to manually add, modify or delete annotation information on the interface, and associates user operation records with corresponding image data for storage for subsequent analysis and tracing.

[0085] Furthermore, in step S8, when the result output module organizes and outputs the results obtained by the target detection and recognition module, the following requirements must be met:

[0086] The report generated by the result output module includes the detailed category of the identified object, the accurate position coordinates in the underwater environment, the three-dimensional size, the shape feature description, the real-time motion status including speed, direction, acceleration, etc., and the motion trend prediction based on the historical trajectory, etc., and also includes the seabed contour map information and the relative position relationship information between the object and the seabed contour, as well as the detailed information of the hazard source, such as the category, location, hazard range, hazard degree, etc., and also includes the recognition accuracy and inferred probability information;

[0087] The charts drawn include a three-dimensional trajectory diagram of the object's position changing over time, a schematic diagram of the shape change under different viewing angles, a curve of the change of motion state parameters, a seabed contour map display, a schematic diagram of the distribution of hazard sources, etc.

[0088] The report also records the performance evaluation indicators of the model, such as target detection accuracy, recall rate, F1 value, position estimation error, motion state prediction error, and hazard source identification accuracy under fusion images, so as to fully understand the analysis effect and model performance.

[0089] Supports storage and export of result data in common formats (such as PDF, CSV, etc.) to facilitate data sharing and interaction with other systems or devices.

[0090] Furthermore, when the image fusion and matching module performs feature matching, if the feature is blurred, partially occluded, or the like, resulting in matching difficulties, a feature matching enhancement method based on deep learning is adopted to learn the semantic information of the features through a pre-trained feature matching network, thereby improving the accuracy and robustness of feature matching, and thereby more accurately identifying the same object in multiple groups of images.

[0091] Furthermore, when analyzing and processing continuous multi-frame fused images, the image analysis module adopts motion estimation technology based on the optical flow method to calculate the motion vector of the object pixel points in the image, assist in determining the object's movement direction and speed, further enrich the object's dynamic feature information, and integrate it into the comprehensive feature vector, thereby enhancing the model's accuracy in identifying the object's motion.

[0092] Furthermore, after the model training is completed, the target detection and recognition module regularly uses the newly acquired image data of collaborative detection by multiple underwater radars to perform incremental learning and optimization on the model. By analyzing the changing trends of the characteristics of the same object in the new data and the characteristics of newly emerging complex scenes, the model structure or parameters are adjusted in a targeted manner, so that the model can continuously adapt to the dynamic changes of the underwater multi-radar detection environment, ensuring that it always has good detection and recognition capabilities for different objects in multi-radar images and accurate identification capabilities for dangerous sources.

[0093] The beneficial effects of the image processing and analysis method for underwater radar signal feedback information of the present invention are as follows: through multi-source signal acquisition and processing, the target detection and recognition accuracy is improved, and the risk of misjudgment and missed judgment is reduced; accurate seabed contour drawing assists marine geology, resource exploration and engineering construction; dynamic target analysis and trajectory prediction serve multiple fields such as vehicle navigation and biological research; intelligent identification and early warning of dangerous sources effectively ensure ship navigation and marine military safety; the user interaction interface is convenient and intuitive, which is convenient for operational decision-making and information traceability management; the result output module promotes data integration and sharing, and promotes multi-faceted coordinated development of the ocean.

[0094] The present invention and its embodiments are described above, which is not restrictive. What is shown in the accompanying drawings is only one of the embodiments of the present invention, and the actual content is not limited to this. In short, if ordinary technicians in this field are inspired by it and do not deviate from the purpose of the invention, they can creatively design structural methods and embodiments similar to the technical scheme, which should all fall within the scope of protection of the present invention.

Claims

1. An image processing and analysis method for underwater radar signal feedback information, characterized in that: The following steps are involved: Step S1: Setting a multi-source signal acquisition module to synchronously acquire the original feedback signal information received after multiple underwater radars transmit signals, and converting them into digital signal formats for storage; Step S2: Establish a signal preprocessing module to perform preliminary processing on the digital signals corresponding to each underwater radar; Step S3: converting the pre-processed underwater radar signals into corresponding image representation forms through an image generation module to obtain multiple groups of underwater radar detection images; Step S4: Setting up an image fusion and matching module to fuse multiple sets of underwater radar detection images and identify the same object in the image, while extracting information related to the seabed contour; Step S5: Setting up an image analysis module to perform comprehensive analysis and processing on the fused images containing the same object and the seabed contour information; Step S6: Setting a target detection and recognition module to perform detailed detection and recognition analysis on the same object based on the processing results of the image analysis module, and identifying the danger sources that may affect the navigation of the ship; Evaluate the recognition accuracy during the detection and recognition process, and make inferences and give corresponding probabilities when the object type cannot be accurately determined, and update the probability as detection continues; Step S7: Establish a user interaction interface module with a labeling function, which is used to label and display the image content on the interface, and visually label and display the results obtained by the target detection and recognition module, including object information, seabed contour information, hazard source information, recognition accuracy, and inference probability information; Step S8: Establish a result output module to organize and output the results obtained by the target detection and recognition module, present them to the user in a visual manner, and support data storage and export functions.

2. The image processing and analysis method for underwater radar signal feedback information according to claim 1, characterized in that: In step S2, when the signal preprocessing module performs preliminary processing on the digital signals corresponding to each underwater radar, the signal preprocessing module includes the following sub-steps: Step S2-1: Use an adaptive bandpass filtering algorithm to remove noise components outside a specific frequency band in each digital signal, dynamically set the filtering frequency band range according to the operating frequency and signal characteristics of each underwater radar, and accurately suppress various types of noise interference; Step S2-2: normalize the denoised digital signals and adjust the signal amplitude according to a unified standard scale to ensure that the images subsequently generated by different underwater radar signals are comparable in terms of intensity, etc., to facilitate subsequent fusion and analysis operations.

3. The image processing and analysis method for underwater radar signal feedback information according to claim 1, characterized in that: In step S4, when the image fusion and matching module fuses multiple groups of underwater radar detection images and identifies the same object corresponding to the image and extracts information related to the seabed contour, it includes the following sub-steps: Step S4-1: spatially register multiple groups of underwater radar detection images. Based on the installation position, attitude information and known underwater geographic coordinates of each underwater radar, objects and seabed contours in different radar images are mapped to the same spatial coordinate system through geometric transformation operations such as coordinate transformation, rotation and translation, laying the foundation for subsequent recognition of the same object and extraction of seabed contours. Step S4-2: Using a feature extraction algorithm to extract features from each registered image, the extracted features include local features of the image, i.e., scale-invariant feature transform (SIFT) features, speeded-up robust features (SURF) features, and global features such as color histogram and texture features, to obtain rich feature description information of each image; Step S4-3: Using a feature matching algorithm, based on the extracted features, find the matching relationship of feature points corresponding to the same object between different images, and by calculating the Euclidean distance between feature vectors as a distance metric, screen out matching pairs that meet the similarity threshold, and determine the areas belonging to the same object in different images based on the matching pairs; Step S4-4: performing fusion processing on the successfully matched regions of the same object, comprehensively considering the information of the object at different viewing angles and different resolutions in each image, and generating a fused object image representation through weighted averaging and multi-resolution fusion methods, so that it contains more comprehensive and accurate object feature information; Step S4-5: For the seabed contour part, the edge detection algorithm is combined with the seabed terrain feature model to extract the seabed contour line from the registered image. By fusing and optimizing the seabed contour information in multiple groups of images, high-precision seabed contour map information is constructed.

4. The image processing and analysis method for underwater radar signal feedback information according to claim 1, characterized in that: In step S5, when the image analysis module performs comprehensive analysis on the fused images containing the same object and the seabed contour information, the following sub-steps are included: Step S5-1: for each frame of fused image, use an image segmentation algorithm to divide the image into different regions, and use a threshold segmentation, region growing or cluster analysis method to preliminarily distinguish possible target regions and background regions; Step S5-2: extract multi-dimensional features from each region, including grayscale features, i.e., mean, variance, and histogram distribution. Texture features are obtained through grayscale co-occurrence matrix and local binary pattern method. Shape features are obtained based on contour analysis and geometric moment calculation to form feature vectors corresponding to each region of each frame image. Step S5-3: Considering the continuous multi-frame fusion image, for the objects in the same position or related area in the image, analyze the feature changes between different frames, calculate the feature difference, change rate and other indicators, such as the grayscale change of the target area between adjacent frames, the change trend of texture complexity, and the deformation degree of the shape, and integrate these feature information reflecting the dynamic changes of the object into the feature vector, and expand it into a comprehensive feature vector that integrates dynamic features; Step S5-4: Correlate and analyze the seabed contour information with the object feature information to determine the relative position relationship between the object and the seabed and the movement trend of the object on the seabed topography, so as to further enrich the information dimension of the comprehensive feature vector.

5. The image processing and analysis method for underwater radar signal feedback information according to claim 1, characterized in that: In step S6, the target detection and recognition module performs detailed detection and recognition analysis on the same object and identifies the hazard source, and evaluates the recognition accuracy, infers the object type probability and updates it, including the following sub-steps: Step S6-1: Construct a comprehensive training data set, collect multiple sets of continuous image data generated by historical collaborative detection of multiple underwater radars, and manually accurately annotate the same object and the corresponding seabed contour information and possible danger sources. The annotation content includes the category of the object, the accurate position, size, shape characteristics and motion state corresponding to different times in the fused image, and other multi-dimensional information. At the same time, the association information between the seabed contour characteristics and the object, as well as the type and degree of danger of the danger source are annotated, so as to construct a training data set covering multi-radar image fusion, object dynamic characteristics, seabed contour association and danger source annotation; Step S6-2: Build a target detection and recognition model based on a deep learning framework, select a network architecture that has the ability to process multi-source data fusion, object dynamic change analysis, seabed contour association, and hazard source identification, such as a multimodal fusion convolutional neural network combined with a recurrent neural network. Use the constructed comprehensive training data set to train the model so that the model learns the mapping relationship between the multi-dimensional features of the same object in the fused image, seabed contour information, hazard source features, and various attributes of the actual target object. Use the stochastic gradient descent optimization algorithm to reasonably adjust the model parameters to improve the accuracy and robustness of the model in object detection and identification and hazard source judgment in multiple radar images and object dynamic scenes; Step S6-3: Input the continuous multi-frame fusion images to be analyzed into the trained target detection and recognition model. The model comprehensively considers the feature information of each frame and the objects between frames, the seabed contour information, and outputs the accurate category, precise location coordinates, detailed size information, shape parameters and real-time motion status of the object such as speed, direction, acceleration, and identifies possible danger sources, including the category, location, danger range and other information of the danger sources; In the output results, the model calculates the recognition accuracy based on the internal confidence evaluation mechanism. When the recognition result is uncertain, it makes inferences about the possible object type and gives the corresponding probability value based on the similarity analysis between the object features and the features of each category in the training data. As new image frames are continuously input for detection, the model constantly updates the inferred probability of object type based on the new feature information; reliable target detection results are screened out based on the preset confidence threshold, achieving accurate analysis of the same object in multiple radar images and effective identification of danger sources.

6. The image processing and analysis method for underwater radar signal feedback information according to claim 1, characterized in that: In the step S7, when the user interaction interface module with annotation function annotates and displays the image content, the following operations are included: Use different colors, shapes or icons to distinguish and mark the identified objects, seabed contours and danger sources on the interface. Use blue lines to outline the seabed contours, green boxes to mark normal objects, and red warning icons to mark danger sources. Provide a detailed display window for the marked information. When the user hovers the mouse over or clicks the marked area, detailed information about the object or hazard source corresponding to the area is displayed, including category, location, size, motion status, degree of danger, recognition accuracy, and inferred probability; It supports users to manually add, modify or delete annotation information on the interface, and associates user operation records with corresponding image data for storage for subsequent analysis and tracing.

7. The image processing and analysis method for underwater radar signal feedback information according to claim 5, characterized in that: In step S8, when the result output module organizes and outputs the results obtained by the target detection and recognition module, the following requirements must be met: The report generated by the result output module includes the detailed category of the identified object, the accurate position coordinates in the underwater environment, the three-dimensional size, the shape feature description, the real-time motion status (including speed, direction, acceleration, etc.) and the motion trend prediction based on the historical trajectory, etc., and also includes the seabed contour map information and the relative position relationship information between the object and the seabed contour, as well as the detailed information of the hazard source, such as the category, location, hazard range, hazard degree, etc., and also includes the recognition accuracy and inferred probability information; The charts drawn include a three-dimensional trajectory diagram of the object's position changing over time, a schematic diagram of the shape change under different viewing angles, a curve of the change of motion state parameters, a seabed contour map display, a schematic diagram of the distribution of hazard sources, etc. The report also records the performance evaluation indicators of the model, including target detection accuracy, recall rate, F1 value, position estimation error, motion state prediction error, and hazard source identification accuracy under fusion images, so as to fully understand the analysis effect and model performance. The result data can be stored and exported in common formats to facilitate data sharing and interaction with other systems or devices.

8. The image processing and analysis method for underwater radar signal feedback information according to claim 3 is characterized in that: When the image fusion and matching module performs feature matching, if the feature is blurred, partially blocked, or the like, which makes matching difficult, a feature matching enhancement method based on deep learning is adopted. The semantic information of the feature is learned through a pre-trained feature matching network to improve the accuracy and robustness of feature matching, thereby more accurately identifying the same object in multiple groups of images.

9. The image processing and analysis method for underwater radar signal feedback information according to claim 4, characterized in that: When analyzing and processing continuous multi-frame fused images, the image analysis module uses motion estimation technology based on the optical flow method to calculate the motion vector of the object pixel points in the image, assist in determining the object's movement direction and speed, further enrich the object's dynamic feature information, and integrate it into the comprehensive feature vector, thereby enhancing the model's accuracy in identifying the object's motion.

10. The image processing and analysis method for underwater radar signal feedback information according to claim 5, characterized in that: After the model training is completed, the target detection and recognition module regularly uses the newly acquired image data of collaborative detection by multiple underwater radars to perform incremental learning and optimization on the model. By analyzing the changing trends of the characteristics of the same object in the new data and the characteristics of newly emerging complex scenes, the model structure or parameters are adjusted in a targeted manner, so that the model can continuously adapt to the dynamic changes of the underwater multi-radar detection environment, ensuring that different objects always have good detection and recognition capabilities under multi-radar images and accurate identification capabilities for dangerous sources.

Images