Dual complementary dynamic convolution optimization method and system for underwater robot image processing

Abstract

The present application provides a dual complementary dynamic convolution optimization method and system for underwater robot image processing, wherein the method includes: in the underwater robot, a polarized laser light source excites a multi-band orthogonal polarized light sequence, and receives a reflected light signal containing the polarization characteristics of suspended object scattering, the polarization modulation characteristics of the reflected light on the surface of the working target, and the characteristics of the water body transmission attenuation spectrum. The main signal frequency band reflected by the working target and the suspended object scattering harmonic interference frequency band are extracted through a noise separation chip, and the input is input into the dual complementary dynamic convolution optimization network to generate an adaptive enhanced feature map linked to the spatial coordinates of the working target. Combined with the optimized feature map of the suspended object scattering harmonic interference frequency band and the water body transmission attenuation spectrum feature, the robot arm trajectory correction parameters and the thruster disturbance compensation coefficient are generated to achieve precise control of the motion trajectory of the underwater robot. The present application improves the target detection accuracy and motion control stability of the underwater robot.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of underwater robot image processing and motion control, and in particular to a dual complementary dynamic convolution optimization method and system for underwater robot image processing. Background Art

[0002] In underwater robot operation scenarios, due to factors such as water turbidity, suspended matter scattering, and complex lighting conditions, target detection and image processing face serious problems of optical signal attenuation and noise interference.

[0003] Currently, one existing solution for underwater image enhancement is a deep learning-based underwater image restoration method. This method trains a convolutional neural network model to learn the distribution characteristics of target reflection signals and suspended object scattering noise directly from images collected in turbid water, and then generates an enhanced image.

[0004] When dealing with complex underwater environments, this solution lacks the ability to model polarized light signals and multi-scale scattering characteristics, making it difficult to effectively distinguish target reflection signals from suspended object scattering noise. This results in serious loss of target details in the enhanced image and limited support for robot motion trajectory control. Summary of the Invention

[0005] The embodiments of the present application provide a dual complementary dynamic convolution optimization method and system for underwater robot image processing, which are used to solve the problems of low target detection accuracy and poor motion control stability of underwater robots in the prior art.

[0006] In a first aspect, the embodiments of the present application provide a dual complementary dynamic convolution optimization method for underwater robot image processing, comprising:

[0007] In the case where a polarized laser light source in an underwater robot excites a multi-band orthogonal polarized light sequence, a reflected light signal and suspended matter concentration gradient distribution data are received, wherein the reflected light signal includes suspended matter scattering polarization characteristics associated with the underwater robot's motion trajectory, polarization modulation characteristics of light reflected from the surface of the operating target, and water body transmission attenuation spectrum characteristics;

[0008] Determine the frequency band of the main signal reflected by the working target according to the polarization modulation characteristics of the reflected light on the surface of the working target, and determine the frequency band of the suspended object scattered harmonic interference according to the polarization characteristics of the suspended object scattered light;

[0009] Input the frequency band of the main signal reflected by the operation target into a preset dual complementary dynamic convolution optimization network, and output an adaptive enhanced feature map linked to the spatial coordinates of the operation target;

[0010] The adaptive enhancement feature map is optimized based on the suspended matter scattering harmonic interference frequency band and the water body transmission attenuation spectrum characteristics to obtain an optimized adaptive enhancement feature map;

[0011] Based on the optimized adaptive enhancement feature map, the trajectory correction parameters of the end effector of the manipulator arm and the disturbance compensation coefficient of the propeller in the underwater robot are generated to control the motion trajectory of the underwater robot.

[0012] Optionally, inputting the frequency band of the main signal reflected by the operation target into a preset dual complementary dynamic convolution optimization network and outputting an adaptive enhanced feature map linked to the spatial coordinates of the operation target includes:

[0013] Obtaining a layered scattering intensity weight and a band modulation phase parameter of the polarized laser light source, wherein the layered scattering intensity weight is generated by a preset multi-scale scattering model according to the suspended matter concentration gradient distribution data;

[0014] generating a set of multi-scale scattering compensation coefficients based on the layered scattering intensity weights and the band modulation phase parameters;

[0015] extracting time-varying polarization modulation features from the motion trajectory data fed back in real time by the underwater robot;

[0016] The time-varying polarization modulation feature and the multi-scale scattering compensation coefficient set are cross-fused in the frequency domain and the spatial domain by a dynamic weight allocation unit in the dual complementary dynamic convolution optimization network to generate a fused feature set;

[0017] In the dual complementary dynamic convolution optimization network, iteratively optimizing the fusion feature set;

[0018] Constructing a three-dimensional geometric constraint field of the target contour based on the main signal frequency band reflected by the operation target and the motion trajectory data;

[0019] Through the feature projection unit in the dual complementary dynamic convolution optimization network, the iteratively optimized fusion feature set is input into the three-dimensional geometric constraint field, and an adaptive enhanced feature map linked to the spatial coordinates of the operation target is output.

[0020] Optionally, the step of inputting the iteratively optimized fusion feature set into the three-dimensional geometric constraint field through the feature projection unit in the dual complementary dynamic convolution optimization network, and outputting an adaptive enhanced feature map linked to the spatial coordinates of the operation target, includes:

[0021] generating an attenuation gradient distribution of a frequency band of a main signal reflected by the operation target;

[0022] The attenuation gradient distribution is coupled with the local deformation parameter of the target surface in a multi-resolution space by a spatial topology adaptation unit in the three-dimensional geometric constraint field to obtain a coupling result;

[0023] In the feature projection unit, based on the coupling result, dynamically correcting the spatial projection path of the iteratively optimized fusion feature set;

[0024] The modified spatial projection path is associated with the propeller disturbance parameters of the underwater robot in real time through a geometric light field synchronization unit in the three-dimensional geometric constraint field to generate a projection path optimization sequence;

[0025] Based on the projection path optimization sequence, the iteratively optimized fusion feature set is subjected to multi-channel feature reorganization according to the geometric constraint relationship of the spatial coordinates in the work target, and an adaptive enhanced feature map is generated that is synchronously linked to the light intensity distribution gradient of the polarization modulation characteristics of the reflected light on the target surface and the acceleration of the robot motion trajectory.

[0026] Optionally, dynamically correcting the spatial projection path of the iteratively optimized fusion feature set based on the coupling result includes:

[0027] Based on the coupling result, cross-scale fusion is performed on the mapping weights between the local deformation parameters of the target surface and the attenuation gradient distribution at different levels of resolution to generate a multi-resolution deformation attenuation correlation tensor;

[0028] According to the dynamic deformation correction coefficient of the three-dimensional geometric constraint field, at the global resolution level, the global projection path direction of the iteratively optimized fusion feature set in the three-dimensional geometric constraint field is corrected by using the displacement gradient direction of the contact surface between the thyroid margin and the carotid artery in the multi-resolution deformation attenuation correlation tensor;

[0029] At the local resolution level, adjusting the local projection path weight of the iteratively optimized fusion feature set in the three-dimensional geometric constraint field by the spatial overlap between the microcalcification cluster density distribution in the multi-resolution deformation attenuated correlation tensor and the metabolomics signal hotspot area;

[0030] Based on the superposition result between the global projection path direction and the local projection path weight, the spatial projection path of the iteratively optimized feature is dynamically corrected.

[0031] Optionally, the time-varying polarization modulation feature and the multi-scale scattering compensation coefficient set are cross-fused in the frequency domain and the spatial domain by a dynamic weight allocation unit in the dual complementary dynamic convolution optimization network to generate a fused feature set, including:

[0032] Performing multi-band decomposition on the time-varying polarization modulation characteristics according to the spatial distribution direction of the underwater robot's motion trajectory to obtain a set of frequency domain orthogonal basis functions;

[0033] The frequency domain orthogonal basis function set and the multi-scale scattering compensation coefficient set are matched in spatial domain convolution kernel scale through the spatial domain feature mapping channel in the dynamic weight allocation unit. During the matching process, the coupling ratio of the frequency domain basis function weights of the frequency domain orthogonal basis function set and the spatial domain convolution kernel scale is dynamically adjusted according to the suspended matter concentration gradient distribution data to generate a frequency domain and spatial domain cross fusion coefficient matrix;

[0034] According to the obstacle distance gradient information in the obstacle avoidance sensor data of the underwater robot, the cross fusion coefficient matrix is ​​dynamically sparsely processed to generate a fusion feature set.

[0035] Optionally, the step of optimizing the adaptive enhancement feature map by combining the suspended matter scattering harmonic interference frequency band and the water body transmission attenuation spectrum characteristics to obtain an optimized adaptive enhancement feature map includes:

[0036] generating a dynamic suppression weight of suspended object scattering noise based on a set of harmonic side lobes in the suspended object scattering harmonic interference frequency band;

[0037] generating a dynamic compensation weight for water body transmission attenuation according to the depth attenuation gradient distribution data in the water body transmission attenuation spectrum characteristics;

[0038] The dynamic suppression weight of the suspended matter scattering noise and the dynamic compensation weight of the water body transmission attenuation are multi-scale fused through the frequency domain and space domain optimization unit in the dual complementary dynamic convolution optimization network to generate a frequency domain and space domain optimization coefficient matrix;

[0039] Based on the frequency domain and spatial domain optimization coefficient matrix, frequency domain compensation and spatial domain enhancement are performed on the target reflection signal frequency band in the adaptive enhancement feature map to generate an optimized adaptive enhancement feature map.

[0040] Optionally, generating the trajectory correction parameters of the end effector of the manipulator and the disturbance compensation coefficient of the propeller in the underwater robot based on the optimized adaptive enhancement feature map includes:

[0041] Extracting target surface local deformation parameters and propeller disturbance harmonic frequency distribution data from the optimized adaptive enhanced feature map;

[0042] According to the local deformation parameters of the target surface, a trajectory correction parameter of the end effector of the manipulator is generated, and based on the propeller disturbance harmonic frequency distribution data, a propeller disturbance compensation coefficient is generated.

[0043] In a second aspect, the embodiments of the present application provide a dual complementary dynamic convolution optimization system for underwater robot image processing, comprising:

[0044] A receiving module is configured to receive reflected light signals and suspended matter concentration gradient distribution data when a polarized laser light source excites a multi-band orthogonal polarized light sequence in the underwater robot. The reflected light signals include suspended matter scattering polarization characteristics associated with the underwater robot's motion trajectory, polarization modulation characteristics of light reflected from the surface of the operating target, and water body transmission attenuation spectrum characteristics.

[0045] A determination module is configured to determine a frequency band of a main signal reflected by the working target according to the polarization modulation characteristics of the light reflected from the surface of the working target, and to determine a frequency band of a harmonic interference wave scattered by the suspended matter according to the polarization characteristics of the suspended matter scattering;

[0046] An input module, configured to input the frequency band of the main signal reflected by the operation target into a preset dual complementary dynamic convolution optimization network, and output an adaptive enhanced feature map linked to the spatial coordinates of the operation target;

[0047] an optimization module, configured to optimize the adaptive enhancement feature map by combining the suspended matter scattering harmonic interference frequency band and the water body transmission attenuation spectrum characteristics to obtain an optimized adaptive enhancement feature map;

[0048] A generation module is used to generate trajectory correction parameters of the end effector of the manipulator arm and disturbance compensation coefficients of the propeller in the underwater robot based on the optimized adaptive enhancement feature map, so as to control the motion trajectory of the underwater robot.

[0049] In a third aspect, an embodiment of the present application provides a computing device comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a dual complementary dynamic convolution optimization method for underwater robot image processing as described in any one of the first aspects.

[0050] In a fourth aspect, an embodiment of the present application provides a computer storage medium having computer program instructions stored thereon, which, when executed by a processor, implements a dual complementary dynamic convolution optimization method for underwater robot image processing as described in any one of the first aspects.

[0051] In an embodiment of the present application, a dual complementary dynamic convolution optimization method for underwater robot image processing is provided. The method includes: receiving a reflected light signal and suspended matter concentration gradient distribution data when a polarized laser light source in the underwater robot excites a multi-band orthogonal polarized light sequence, the reflected light signal including a suspended matter scattering polarization characteristic, a polarization modulation characteristic of light reflected from a target surface, and a water body transmission attenuation spectrum characteristic associated with the underwater robot's motion trajectory; determining a main signal frequency band reflected from the target surface based on the polarization modulation characteristic of light reflected from the target surface, and determining a suspended matter scattering harmonic interference frequency band based on the suspended matter scattering polarization characteristic; inputting the main signal frequency band reflected from the target into a preset dual complementary dynamic convolution optimization network, outputting an adaptive enhancement feature map linked to the spatial coordinates of the target; optimizing the adaptive enhancement feature map based on the suspended matter scattering harmonic interference frequency band and the water body transmission attenuation spectrum characteristic to obtain an optimized adaptive enhancement feature map; and generating trajectory correction parameters for an end effector of a manipulator arm and a disturbance compensation coefficient for a thruster in the underwater robot based on the optimized adaptive enhancement feature map to control the motion trajectory of the underwater robot.

[0052] The technical solution of this application has the following beneficial effects:

[0053] This application uses a polarized laser light source to excite a multi-band orthogonal polarized light sequence to obtain a mixed light signal containing the target reflection signal and suspended object scattering noise, providing high-quality input data for subsequent signal separation and image enhancement. The target reflection signal and the suspended object scattering noise are accurately separated by utilizing the polarization modulation characteristics of the target surface reflected light and the polarization characteristics of the suspended object scattering, laying the foundation for image enhancement. The target reflection main signal frequency band is processed by the dual complementary dynamic convolution optimization network to generate an adaptive enhancement feature map linked to the spatial coordinates of the operating target, thereby improving the target detection accuracy. Combined with the suspended object scattering harmonic interference frequency band and the water body transmission attenuation spectrum characteristics, the feature map is further optimized to suppress noise interference and enhance target details. Based on the optimized adaptive enhancement feature map, the robot arm trajectory correction parameters and the thruster disturbance compensation coefficient are generated to achieve precise control of the underwater robot motion trajectory.

[0054] Furthermore, the embodiment of the present application also generates layered scattering intensity weights through a multi-scale scattering model, and generates a multi-scale scattering compensation coefficient set in combination with the band modulation phase parameters of the polarized laser light source; extracts time-varying polarization modulation features from the underwater robot motion trajectory data, and uses a dynamic weight allocation unit to cross-fuse it with the scattering compensation coefficient set in the frequency domain and spatial domain to generate a fusion feature set; in a three-dimensional geometric constraint field, the iteratively optimized fusion feature set is mapped into an adaptive enhanced feature map linked to the spatial coordinates of the operation target through a feature projection unit.

[0055] Moreover, through multi-scale scattering compensation and frequency domain and spatial domain cross-fusion, the influence of suspended matter scattering noise and water body transmission attenuation on the target reflection signal can be effectively suppressed, and a high-quality adaptive enhanced feature map can be generated, providing accurate spatial coordinate information and environmental perception data for underwater robot target detection and motion control.

[0056] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0058] Figure 1 A flowchart of a dual complementary dynamic convolution optimization method for underwater robot image processing provided in an embodiment of the present application;

[0059] Figure 2 A schematic diagram of the structure of a dual complementary dynamic convolution optimization system for underwater robot image processing provided by an embodiment of the present application;

[0060] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0061] In order to enable people skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0062] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0063] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0064] To address the problems of low target detection accuracy and severe image noise interference in underwater robot operation scenarios, this solution uses a polarized laser light source to excite a multi-band orthogonal polarized light sequence, receives reflected light signals containing the polarization characteristics of suspended matter scattering, the polarization modulation characteristics of light reflected from the surface of the operation target, and the characteristics of the water body transmission attenuation spectrum, and uses a noise separation chip to extract the main signal frequency band of the operation target reflection and the suspended matter scattering harmonic interference frequency band; inputs the target reflection main signal frequency band into the dual complementary dynamic convolution optimization network to generate an adaptive enhanced feature map linked to the spatial coordinates of the operation target; combines the suspended matter scattering harmonic interference frequency band and the water body transmission attenuation spectrum characteristic optimization feature map to generate the robot arm trajectory correction parameters and thruster disturbance compensation coefficient, realize the precise control of the underwater robot motion trajectory, and thus improve the target detection accuracy and motion stability.

[0065] Figure 1 A flowchart of a dual complementary dynamic convolution optimization method for underwater robot image processing provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes:

[0066] Step 101: When a polarized laser light source in an underwater robot excites a multi-band orthogonal polarized light sequence, a reflected light signal and suspended matter concentration gradient distribution data are received.

[0067] In this step, the polarized laser light source is a light source capable of emitting a multi-band sequence of orthogonally polarized light, which is used to stimulate reflected light signals from the target surface and suspended objects. The reflected light signal contains the suspended object scattering polarization characteristics associated with the underwater robot's motion trajectory, the polarization modulation characteristics of the light reflected from the target surface, and the water body transmission attenuation spectrum characteristics, which are used for subsequent signal separation and image enhancement. Suspended matter refers to tiny particles or substances suspended in the water, including silt, plankton, and organic debris. These particles scatter and attenuate light propagation. The target is the specific object that the underwater robot needs to detect, identify, locate, or manipulate during its mission, such as a submarine pipeline, a sunken ship, or a biological sample. The water body refers to the aquatic medium in the underwater robot's operating environment, including the water itself and dissolved or suspended substances therein, such as suspended matter, plankton, and dissolved gases. Suspended matter concentration gradient distribution data refers to the distribution data describing the variation of suspended matter concentration with spatial position in the underwater environment and is used to analyze the suspended matter scattering characteristics.

[0068] In this embodiment, a polarized laser light source emits a multi-band sequence of orthogonally polarized light. After irradiating the underwater environment and the target surface, it receives reflected light signals and suspended solids concentration gradient distribution data. The reflected light signals contain the polarization characteristics of suspended solids scattering, the polarization modulation characteristics of light reflected from the target surface, and the water transmission attenuation spectrum. These characteristics are collected in real time by an optical sensor and transmitted to a noise separation chip. The suspended solids concentration gradient distribution data is collected by an obstacle avoidance sensor and used as input for the subsequent multi-scale scattering model.

[0069] For example, during an underwater archaeological mission, an underwater robot was deployed to locate a shipwreck. First, it scanned the seafloor using polarized laser light, collecting reflected light signals. By analyzing these signals, it not only successfully identified a potential target, which turned out to be a shipwreck, but also accurately measured the distribution of surrounding suspended matter, providing critical data for subsequent steps.

[0070] Step 102: determining a frequency band of a main signal reflected by the working target according to the polarization modulation characteristics of the light reflected from the working target surface, and determining a frequency band of suspended matter scattered harmonic interference according to the suspended matter scattered polarization characteristics.

[0071] In this step, the main target reflection signal frequency band refers to the frequency band extracted from the reflected light signal dominated by the target's reflected light. It contains information about the target's surface material, roughness, and geometry. The polarization modulation feature of the target's surface reflected light is a key feature extracted from the reflected light signal, used to distinguish the target's reflected light from light scattered by suspended matter. This dominant frequency band of the target's reflected light, derived by decoupling the reflected light signal based on the polarization modulation feature, serves as the core input for image processing. The suspended matter scattering harmonic interference frequency band refers to the frequency band extracted from the reflected light signal dominated by suspended matter scattered light, manifesting as a harmonic interference signal. The suspended matter concentration gradient distribution data directly influences the characteristics of the suspended matter scattering harmonic interference frequency band. Specifically, higher suspended matter concentration increases the scattered light intensity and the energy of the harmonic interference frequency band. The spatial gradient of the suspended matter distribution (i.e., concentration variation) causes changes in the frequency components of the scattered light, thereby affecting the spectral characteristics of the harmonic interference frequency band.

[0072] In this embodiment, a noise separation chip performs polarization-time-frequency decoupling on the reflected light signal to extract the main signal band reflected by the target and the harmonic interference band of suspended matter scattering. The main signal band is extracted based on the phase offset in the polarization modulation characteristics of the light reflected from the target surface, while the harmonic interference band of suspended matter scattering is extracted based on the random polarization states in the polarization characteristics of suspended matter scattering.

[0073] For example, continuing with the previous scenario, in the acquired reflected light signal, the system automatically identifies the unique reflection pattern on the surface of the shipwreck as the main signal frequency band, and at the same time marks the interference frequency band caused by the surrounding suspended objects, providing a basis for the next step of accurately locating the shipwreck.

[0074] Step 103: Input the frequency band of the main signal reflected by the operation target into a preset dual complementary dynamic convolution optimization network, and output an adaptive enhanced feature map linked to the spatial coordinates of the operation target.

[0075] In this step, the dual complementary dynamic convolutional optimization network refers to a network structure used for image enhancement. It generates an adaptive enhanced feature map by alternating between spatial domain feature focusing and channel domain feature compensation operations. This linkage is reflected in the dynamic correlation between the generation process of the adaptive enhanced feature map and the spatial coordinates, motion state, task requirements, and environmental changes of the underwater robot's operating target. This linkage mechanism ensures a high degree of consistency between the image processing results and the robot's operating target, thereby improving the accuracy and efficiency of underwater robot operations. The adaptive enhanced feature map is an enhanced image that is linked to the spatial coordinates of the operating target and contains target contour and texture detail information.

[0076] In an embodiment of the present application, the frequency band of the main signal reflected by the work target is input into the dual complementary dynamic convolution optimization network, and an adaptive enhanced feature map linked to the spatial coordinates of the work target is generated by alternately activating the radial constraint mode of the spatial domain convolution kernel and the scattering compensation mode of the channel domain convolution kernel.

[0077] For example, based on the data obtained in the previous step, the system inputs the main reflection frequency band of the shipwreck into the dual complementary dynamic convolution optimization network, generating a detailed structural map of the shipwreck. This map clearly marks the various parts of the shipwreck, improving the efficiency of subsequent exploration.

[0078] Step 104: Optimizing the adaptive enhancement feature map based on the suspended matter scattering harmonic interference frequency band and the water body transmission attenuation spectrum characteristics to obtain an optimized adaptive enhancement feature map.

[0079] In this step, the optimized adaptive enhancement feature map refers to an enhancement feature map that further improves image quality by suppressing suspended matter scattering noise and compensating for water body transmission attenuation characteristics.

[0080] In an embodiment of the present application, the adaptive enhancement feature map is optimized by combining the suspended matter scattering harmonic interference frequency band and the water body transmission attenuation spectrum characteristics. Through multi-scale scattering compensation and frequency domain and spatial domain cross-fusion, the suspended matter scattering noise is suppressed and the target details are enhanced to generate an optimized adaptive enhancement feature map.

[0081] For example, based on previous shipwreck images, the system comprehensively considered the impact of surrounding suspended matter and the optical properties of seawater, and corrected the shipwreck structure map, eliminating the distortion caused by environmental factors and making the outline of the shipwreck more clearly visible.

[0082] Step 105: Based on the optimized adaptive enhancement feature map, the trajectory correction parameters of the end effector of the manipulator and the disturbance compensation coefficient of the propeller in the underwater robot are generated to control the motion trajectory of the underwater robot.

[0083] In this step, the trajectory correction parameters are used to adjust the parameters of the robot arm's end effector's motion path to ensure that the robot arm accurately approaches or operates the target. The disturbance compensation coefficient is used to suppress the effect of thruster disturbances on the suspended object scattering noise, thereby optimizing the thruster's working state.

[0084] In an embodiment of the present application, based on the optimized adaptive enhancement feature map, the target contour geometric constraint data is extracted to generate the robot arm trajectory correction parameters; combined with the suspended object scattering noise suppression weight, the thruster disturbance compensation coefficient is generated to achieve precise control of the underwater robot's motion trajectory.

[0085] For example, in the end, the underwater robot adjusted its own route and posture based on the optimized shipwreck structure diagram, successfully approached the shipwreck and carried out detailed investigation work.

[0086] This scheme uses a polarized laser light source to excite a multi-band orthogonal polarized light sequence, receives and separates the target reflection signal and the suspended object scattering noise, and uses a dual complementary dynamic convolution optimization network to generate an adaptive enhanced feature map linked to the spatial coordinates of the operating target. It further optimizes the feature map and generates the robot arm trajectory correction parameters and thruster disturbance compensation coefficients, thereby improving the underwater robot's target detection accuracy and motion control stability.

[0087] In order to solve the problems of low target detection accuracy, severe image noise interference, and unstable motion control of underwater robots, in some embodiments, step 103: inputting the frequency band of the main signal reflected by the operating target into a preset dual complementary dynamic convolution optimization network, and outputting an adaptive enhanced feature map linked to the spatial coordinates of the operating target, includes:

[0088] Step 201: Obtaining the layered scattering intensity weight and the band modulation phase parameter of the polarized laser light source.

[0089] In step 201, the layered scattering intensity weight is a parameter generated by a preset multi-scale scattering model based on the suspended matter concentration gradient distribution data. It is used to describe the contribution of suspended matter of different particle sizes to light scattering. The band modulation phase parameter is a parameter of the polarized laser light source that is dynamically adjusted based on environmental conditions and mission requirements to optimize the emission characteristics of the optical signal.

[0090] In this embodiment, a multiscale scattering model is used to analyze suspended matter concentration gradient distribution data to generate layered scattering intensity weights. Simultaneously, the polarized laser light source dynamically adjusts the band modulation phase parameters based on current environmental conditions. These parameters are used for subsequent multiscale scattering compensation and feature fusion.

[0091] Step 202: Generate a set of multi-scale scattering compensation coefficients based on the layered scattering intensity weights and the band modulation phase parameters.

[0092] In step 202, a multi-scale scattering compensation coefficient set is used to compensate for the influence of suspended object scattering at different scales on the target reflection signal, thereby ensuring the authenticity and accuracy of the reflection signal.

[0093] In an embodiment of the present application, based on the layered scattering intensity weight, a dynamic compensation relationship between scattering characteristics of different scales in the frequency band of the main signal reflected by the working target is established, and a set of multi-scale scattering compensation coefficients is generated according to the dynamic supplementary relationship combined with the band modulation phase parameter.

[0094] Step 203: extracting time-varying polarization modulation features from the motion trajectory data fed back in real time by the underwater robot.

[0095] In step 203, the time-varying polarization modulation feature refers to the change in polarization light characteristics caused by the movement of the underwater robot over time.

[0096] In this embodiment, the time-varying polarization modulation features are extracted by analyzing the propeller speed fluctuation parameters and the manipulator joint angular acceleration data in the underwater robot's motion trajectory data. These features are used to describe the dynamic impact of the robot's motion on the polarization state of the target reflection signal.

[0097] Step 204: The time-varying polarization modulation feature and the multi-scale scattering compensation coefficient set are cross-fused in the frequency domain and the spatial domain by the dynamic weight allocation unit in the dual complementary dynamic convolution optimization network to generate a fused feature set.

[0098] In step 204, the dynamic weight allocation unit, a functional unit within the dual-complementary dynamic convolutional optimization network, is used to perform frequency-domain and spatial-domain cross-fusion of the time-varying polarization modulation features and the multi-scale scattering compensation coefficient set. The fused feature set is an intermediate feature set generated through frequency-domain and spatial-domain cross-fusion and is used for subsequent iterative optimization.

[0099] In this embodiment of the present application, a dynamic weight allocation unit is used to cross-fuse the time-varying polarization modulation features with the multi-scale scattering compensation coefficient set in the frequency and spatial domains to generate a fused feature set. The fusion process includes frequency domain basis function weight adjustment and spatial domain convolution kernel scale matching.

[0100] Step 205: In the dual complementary dynamic convolution optimization network, iteratively optimize the fusion feature set.

[0101] In an embodiment of the present application, in the dual complementary dynamic convolution optimization network, the fusion feature set is iteratively optimized by alternately activating the radial constraint mode of the spatial domain convolution kernel and the scattering compensation mode of the channel domain convolution kernel; wherein the radial constraint mode dynamically adjusts the topological structure generation of the convolution kernel according to the polarization modulation characteristics of the reflected light on the target surface, and the scattering compensation mode dynamically corrects the feature transfer weight generation between channels according to the suspended matter concentration gradient distribution data.

[0102] Step 206 constructs a three-dimensional geometric constraint field of the target profile based on the frequency band of the main signal reflected by the work target and the motion trajectory data.

[0103] In step 206, the target profile refers to the geometric shape and spatial structure information corresponding to the polarization modulation characteristics of the reflected light on the surface of the work target. The three-dimensional geometric constraint field is a mathematical model used to describe the spatial position and shape characteristics of the target object.

[0104] In the embodiment of the present application, a three-dimensional geometric constraint field is constructed based on the target contour geometric constraint data in the frequency band of the main signal reflected by the operating target and the underwater robot motion trajectory data. This constraint field is used to guide the generation of feature projection and adaptive enhancement feature maps.

[0105] Step 207: The iteratively optimized fusion feature set is input into the three-dimensional geometric constraint field through the feature projection unit in the dual complementary dynamic convolution optimization network, and an adaptive enhanced feature map linked to the spatial coordinates of the operation target is output.

[0106] In step 207, the feature projection unit refers to a functional unit in the dual complementary dynamic convolution optimization network, which is used to map the iteratively optimized fusion feature set into a three-dimensional geometric constraint field.

[0107] In an embodiment of the present application, a feature projection unit is used to map the optimized fusion feature set into a three-dimensional geometric constraint field to generate an adaptive enhanced feature map that is updated synchronously with the spatial coordinates of the operation target.

[0108] Here's a specific example:

[0109] During an underwater archaeological mission, an underwater robot first collected concentration gradient distribution data of surrounding suspended matter and generated layered scattering intensity weights and band modulation phase parameters based on this data. The robot then monitored its own trajectory in real time and extracted time-varying polarization modulation features from this data. These features were then fused with a previously acquired set of compensation coefficients. This fused feature set was then used as input for iterative optimization within a dual-complementary dynamic convolutional optimization network. Based on the optimization results and the target reflection signal, a detailed three-dimensional geometric constraint field was constructed. Finally, a feature projection unit was used to map the optimized feature set onto this constraint field, resulting in a clear image of the target.

[0110] This solution improves the underwater robot's target detection accuracy and image quality through multi-scale scattering compensation, cross-fusion of frequency domain and spatial domain, and the construction of a three-dimensional geometric constraint field. At the same time, by generating robotic arm trajectory correction parameters and thruster disturbance compensation coefficients, it achieves precise control of the robot's motion trajectory, effectively solving the difficult problems of target detection and motion control in underwater environments.

[0111] In order to further improve the linkage accuracy between the adaptive enhanced feature map and the spatial coordinates of the operation target, in some embodiments, step 207: inputting the iteratively optimized fusion feature set into the three-dimensional geometric constraint field through the feature projection unit in the dual complementary dynamic convolution optimization network, and outputting the adaptive enhanced feature map linked to the spatial coordinates of the operation target, includes:

[0112] Step 301: Generate an attenuation gradient distribution of a frequency band of a main signal reflected by the operation target.

[0113] In step 301, the attenuation gradient distribution is used to describe the distribution data of the attenuation characteristics of the main signal frequency band reflected by the operation target at different water depths, so as to guide the correction of the spatial projection path.

[0114] In this embodiment, the attenuation gradient distribution of the target reflected signal at different water depths is calculated based on the light intensity distribution data in the frequency band of the main signal reflected by the target. This data is used for subsequent multi-resolution spatial coupling and projection path correction.

[0115] Step 302: The attenuation gradient distribution is coupled with the local deformation parameters of the target surface in a multi-resolution spatial manner through a spatial topology adaptation unit in the three-dimensional geometric constraint field to obtain a coupling result.

[0116] In step 302, the spatial topology adaptation unit refers to a functional unit in the three-dimensional geometric constraint field, which performs multi-resolution spatial coupling on the attenuation gradient distribution and the local deformation parameters of the target surface. The coupling result refers to the intermediate result generated by multi-resolution spatial coupling, which is used to guide the correction of the spatial projection path. The local deformation parameters are parameters used to quantify the local changes in the geometric shape of the working target surface, reflecting the geometric characteristics of the target surface such as curvature, convexity and edge. The local deformation parameters are extracted from the polarization modulation characteristics of the reflected light on the working target surface. The geometric shape of the target surface will affect the polarization state of the reflected light, thereby forming a specific polarization modulation feature.

[0117] In the embodiment of the present application, a spatial topology adaptation unit is used to perform multi-resolution spatial coupling between the attenuation gradient distribution and the local deformation parameters of the target surface to generate a coupling result. The coupling process includes attenuation compensation at the global resolution level and deformation correction at the local resolution level.

[0118] Step 303: In the feature projection unit, based on the coupling result, dynamically correct the spatial projection path of the iteratively optimized fusion feature set.

[0119] In step 303, the correction process of the spatial projection path includes: adjusting the radiation attenuation compensation weight of the characteristic projection according to the suspended matter concentration gradient distribution data, and synchronously updating the polarization sensitivity threshold of the projection path according to the band modulation phase parameter of the polarized laser light source.

[0120] In an embodiment of the present application, in a feature projection unit, a spatial projection path is dynamically corrected for the iteratively optimized fusion feature set based on the coupling result. The correction process includes adjusting the radiation attenuation compensation weight and polarization sensitivity threshold of the projection path.

[0121] Step 304: The corrected spatial projection path is associated with the propeller disturbance parameters of the underwater robot in real time through the geometric light field synchronization unit in the three-dimensional geometric constraint field to generate a projection path optimization sequence.

[0122] In step 304, the geometric light field synchronization unit, a functional unit within the 3D geometric constraint field, is responsible for associating the corrected spatial projection path with the propeller disturbance parameters in real time. The propeller disturbance parameters are input data describing the characteristics of the propeller disturbance; the propeller disturbance compensation coefficients are output data used to mitigate the effects of the propeller disturbance. The projection path optimization sequence, generated by the geometric light field synchronization unit, guides the final generation of the feature map.

[0123] In this embodiment, the geometric light field synchronization unit associates the corrected spatial projection path with the underwater robot's propeller disturbance parameters in real time to generate projection path optimization sequences. These sequences are used to ensure the synchronization of the feature map generation process and the robot's motion state.

[0124] Step 305: Based on the projection path optimization sequence, the iteratively optimized fusion feature set is subjected to multi-channel feature reorganization according to the geometric constraint relationship of the spatial coordinates in the work target, and an adaptive enhanced feature map is generated that is synchronized with the light intensity distribution gradient of the polarization modulation characteristics of the reflected light on the target surface and the acceleration of the robot motion trajectory.

[0125] In step 305, multi-channel feature recombination refers to recombining the iteratively optimized fusion feature set according to the geometric constraint relationship of the operation target space coordinates to generate a final adaptive enhanced feature map.

[0126] In an embodiment of the present application, based on the projection path optimization sequence, the feature set is reorganized according to the spatial coordinates and geometric constraints, and finally an adaptive enhanced feature map is generated that can simultaneously reflect the target surface reflection light characteristics, light intensity distribution gradient and robot motion trajectory.

[0127] Here's a specific example:

[0128] During an underwater archaeological mission, an underwater robot first analyzed the light signal reflected from the target to generate an attenuation gradient distribution. This distribution was then combined with the target surface deformation parameters using a spatial topology adaptation unit to obtain a coupled result. Based on this result, the fused feature set was dynamically corrected to the spatial projection path. Subsequently, the geometric light field synchronization unit linked the corrected path with the propeller perturbation parameters to generate an optimized projection path sequence. Finally, the feature set was reorganized into multiple channels according to the optimized sequence and spatial coordinate geometry constraints, generating a clear and accurate target image that also reflected the dynamic changes in the robot's motion trajectory.

[0129] This solution improves the linkage accuracy between the adaptive enhancement feature map and the spatial coordinates of the operating target through multi-resolution spatial coupling, dynamic correction of the spatial projection path, and geometric light field synchronization. At the same time, it generates high-quality feature maps through multi-channel feature recombination, providing accurate environmental perception data for underwater robot target detection and motion control.

[0130] In order to further improve the accuracy and adaptability of the spatial projection path correction, in some embodiments, step 303: dynamically correcting the spatial projection path of the iteratively optimized fusion feature set based on the coupling result includes:

[0131] Step 401: Based on the coupling result, the mapping weights between the local deformation parameters of the target surface at different levels of resolution and the attenuation gradient distribution are cross-scale fused to generate a multi-resolution deformation attenuation correlation tensor.

[0132] In step 401, the multi-resolution deformation attenuation correlation tensor refers to a tensor generated by cross-scale fusion of the local deformation parameters of the target surface and the attenuation gradient distribution, which is used to describe the correlation between deformation and attenuation at different resolution levels.

[0133] In an embodiment of the present application, a multi-resolution deformation attenuation correlation tensor is a tensor generated by cross-scale fusion of local deformation parameters of the target surface and the attenuation gradient distribution, which is used to describe the correlation between deformation and attenuation at different resolution levels.

[0134] Step 402: Based on the dynamic deformation correction coefficient of the three-dimensional geometric constraint field, at the global resolution level, the global projection path direction of the iteratively optimized fusion feature set in the three-dimensional geometric constraint field is corrected through the displacement gradient direction of the contact surface between the thyroid edge and the carotid artery in the multi-resolution deformation attenuation correlation tensor.

[0135] In step 402, the global resolution level and local resolution level are used to describe the processing methods of different resolution levels in the multi-scale spatial coupling process. The global resolution level refers to the resolution level that comprehensively analyzes and processes the target surface geometric deformation and optical signal propagation characteristics at a larger scale. The global projection path direction refers to the projection path direction corrected at the global resolution level, which is used to ensure the consistency of the feature map generation process with the overall geometric characteristics of the target.

[0136] In an embodiment of the present application, based on the dynamic deformation correction coefficient of the three-dimensional geometric constraint field, at the global resolution level, the global projection path direction of the iteratively optimized fusion feature set in the three-dimensional geometric constraint field is corrected by the displacement gradient direction of the target edge and the contact surface in the multi-resolution deformation attenuation correlation tensor.

[0137] Step 403: At the local resolution level, the local projection path weight of the iteratively optimized fusion feature set in the three-dimensional geometric constraint field is adjusted based on the spatial overlap between the microcalcification cluster density distribution in the multi-resolution deformation attenuated correlation tensor and the metabolomics signal hotspot area.

[0138] In step 403, the local resolution level refers to the resolution level at which fine-grained analysis and processing of the target surface geometry and optical signal propagation characteristics are performed at a smaller scale. The local projection path weight refers to the projection path weight adjusted at the local resolution level to enhance the representation of local target details in the feature map.

[0139] In an embodiment of the present application, at the local resolution level, the local projection path weight of the iteratively optimized fusion feature set in the three-dimensional geometric constraint field is adjusted by adjusting the spatial overlap between the microcalcification cluster density distribution in the multi-resolution deformation attenuated correlation tensor and the metabolomics signal hotspot area.

[0140] Step 404: Based on the superposition result between the global projection path direction and the local projection path weight, dynamically correct the spatial projection path of the iteratively optimized feature.

[0141] In an embodiment of the present application, based on the superposition result between the global projection path direction and the local projection path weight, the spatial projection path of the iteratively optimized fusion feature set is dynamically corrected to ensure that the feature map generation process accurately matches the target geometric characteristics and environmental conditions.

[0142] Here's a specific example:

[0143] In an underwater archaeological mission, the underwater robot first extracted the deformation parameters and signal attenuation of the shipwreck hull from the multi-scale scattering model and reflected light signal, generating a multi-resolution deformation attenuation correlation tensor. Then, using the information about the shipwreck hull and the surrounding coral cover in this tensor, combined with the dynamic deformation correction coefficient, the global projection path direction was adjusted. Next, by analyzing the spatial overlap between the density distribution of artifact fragments and the thickness of the sediment layer, the local projection path weights were adjusted to more accurately locate small artifact fragments. Finally, the results of the global and local adjustments were combined to dynamically correct the spatial projection path of the fused features, thereby obtaining a more accurate representation of the target features.

[0144] In order to further improve the accuracy and adaptability of the cross-fusion of frequency domain and spatial domain, in some embodiments, step 204: the dynamic weight allocation unit in the dual complementary dynamic convolution optimization network is used to cross-fuse the time-varying polarization modulation feature with the multi-scale scattering compensation coefficient set in the frequency domain and spatial domain to generate a fused feature set, including:

[0145] Step 501: performing multi-band decomposition on the time-varying polarization modulation characteristics according to the spatial distribution direction of the underwater robot's motion trajectory to obtain a set of frequency-domain orthogonal basis functions.

[0146] In step 501, the frequency domain orthogonal basis function set is a group of independent basis functions obtained by decomposing the time-varying polarization modulation characteristics according to different frequency components, and is used to represent different frequency characteristics of the signal.

[0147] In this embodiment, the underwater robot's trajectory data is first analyzed to determine its spatial distribution direction. Frequency-domain analysis techniques, such as Fourier transforms, are then used to perform a multi-band decomposition of the time-varying polarization modulation signature, extracting a series of frequency-domain orthogonal basis functions that represent the signal characteristics at different frequency components.

[0148] Step 502: Through the spatial feature mapping channel in the dynamic weight allocation unit, the frequency domain orthogonal basis function set and the multi-scale scattering compensation coefficient set are matched in the spatial domain convolution kernel scale. During the matching process, the coupling ratio of the frequency domain basis function weights of the frequency domain orthogonal basis function set and the spatial domain convolution kernel scale is dynamically adjusted according to the suspended matter concentration gradient distribution data to generate a frequency domain and spatial domain cross-fusion coefficient matrix.

[0149] In step 502, the spatial feature mapping channel refers to a functional module within the dynamic weight allocation unit that is used to perform spatial convolution kernel scale matching between the set of frequency-domain orthogonal basis functions and the set of multi-scale scattering compensation coefficients. The frequency-domain and spatial-domain cross-fusion coefficient matrix is ​​a matrix generated by dynamically adjusting the coupling ratio between the frequency-domain basis function weights and the spatial-domain convolution kernel scales, and is used to guide the generation of the fused feature set.

[0150] In this embodiment, the spatial feature mapping channel in the dynamic weight allocation unit is used to match a set of frequency-domain orthogonal basis functions with a set of multi-scale scattering compensation coefficients. During this process, the ratio of the frequency-domain basis function weights to the spatial convolution kernel scale is adjusted in real time based on the suspended matter concentration gradient distribution data, ultimately generating a cross-fusion coefficient matrix that incorporates both frequency-domain and spatial-domain information.

[0151] Step 503: According to the obstacle distance gradient information in the obstacle avoidance sensor data of the underwater robot, the cross fusion coefficient matrix is ​​dynamically sparsely processed to generate a fusion feature set.

[0152] In step 503, the obstacle distance gradient information refers to data collected in real time by the underwater robot's obstacle avoidance sensor, describing the rate of change of the obstacle distance around the robot.

[0153] In this embodiment, a dynamic sparsification operation is performed on the cross-fusion coefficient matrix based on the obstacle distance gradient information provided by the underwater robot's obstacle avoidance sensor. This process involves using an algorithm to identify and reduce unimportant features, retaining those that are critical for target recognition, and thus generating an optimized fused feature set.

[0154] Here's a specific example:

[0155] During an underwater archaeological mission, an underwater robot first performed a multi-band decomposition of the time-varying polarization modulation characteristics of its trajectory, obtaining a set of frequency-domain orthogonal basis functions reflecting different frequency characteristics. Next, using the spatial feature mapping channel within the dynamic weight allocation unit and combined with suspended matter concentration gradient distribution data, these basis functions were matched with a set of multi-scale scattering compensation coefficients to generate a frequency-domain and spatial-domain cross-fusion coefficient matrix. Finally, this matrix was dynamically sparsified based on obstacle distance information provided by the obstacle avoidance sensor, ensuring that only the key features most conducive to target identification were retained, forming the final fused feature set.

[0156] This scheme improves the accuracy and adaptability of frequency domain and spatial domain cross-fusion through frequency domain orthogonal basis function decomposition, spatial domain convolution kernel scale matching and dynamic sparsification processing, generates a high-quality fusion feature set, and provides accurate environmental perception data for underwater robot target detection and motion control.

[0157] In order to further improve the image quality and target detection accuracy of the adaptive enhancement feature map, in some embodiments, step 104: optimizing the adaptive enhancement feature map by combining the suspended matter scattering harmonic interference frequency band and the water body transmission attenuation spectrum characteristics to obtain the optimized adaptive enhancement feature map includes:

[0158] Step 601: Generate dynamic suppression weights for suspended object scattering noise based on a set of harmonic side lobes in the suspended object scattering harmonic interference frequency band.

[0159] In step 601, the harmonic sidelobe set refers to the harmonic sidelobe components extracted from the suspended object scattering harmonic interference frequency band, and is used to describe the spectrum distribution characteristics of the suspended object scattering noise. The dynamic suppression weight is a parameter used to reduce the impact of suspended object scattering noise.

[0160] In the embodiment of the present application, based on the harmonic sidelobe set in the suspended matter scattering harmonic interference frequency band, the spectrum distribution characteristics of the suspended matter scattering noise are extracted to generate dynamic suppression weights. These weights are used for subsequent multi-scale fusion and noise suppression.

[0161] Step 602: Generate a dynamic compensation weight for water body transmission attenuation based on the depth attenuation gradient distribution data in the water body transmission attenuation spectrum characteristics.

[0162] In step 602, the depth attenuation gradient distribution data refers to the gradient distribution data describing the variation of water transmission attenuation with depth, which is used to analyze the propagation characteristics of light signals in water. The dynamic compensation weight is a parameter used to correct this attenuation effect.

[0163] In this embodiment of the present application, the frequency domain compensation coefficients of water body transmission attenuation are extracted based on the depth attenuation gradient distribution data in the water body transmission attenuation spectrum characteristics to generate dynamic compensation weights. These weights are used for subsequent multi-scale fusion and attenuation compensation.

[0164] Step 603: The dynamic suppression weight of the suspended matter scattering noise and the dynamic compensation weight of the water body transmission attenuation are multi-scale fused through the frequency domain and spatial domain optimization unit in the dual complementary dynamic convolution optimization network to generate a frequency domain and spatial domain optimization coefficient matrix.

[0165] In step 603, the frequency-domain and spatial-domain optimization unit is a functional module within the dual-complementary dynamic convolutional optimization network, which is used to perform multi-scale fusion of dynamic suppression weights and dynamic compensation weights. The frequency-domain and spatial-domain optimization coefficient matrix is ​​used to guide the frequency-domain compensation and spatial-domain enhancement of the adaptive enhancement feature map.

[0166] In this embodiment, a frequency-domain and spatial-domain optimization unit is used to perform a multi-scale fusion of the dynamic suppression weights for suspended solids scattering noise and the dynamic compensation weights for water transmission attenuation to generate a frequency-domain and spatial-domain optimization coefficient matrix. The fusion process includes frequency-domain basis function weight adjustment and spatial-domain convolution kernel scale matching.

[0167] Step 604: Based on the frequency domain and spatial domain optimization coefficient matrix, frequency domain compensation and spatial domain enhancement are performed on the target reflection signal frequency band in the adaptive enhancement feature map to generate an optimized adaptive enhancement feature map.

[0168] In step 604, frequency domain compensation and spatial domain enhancement refer to improving the feature map in frequency and spatial dimensions respectively to enhance its quality and accuracy.

[0169] In this embodiment of the present application, based on the frequency domain and spatial domain optimization coefficient matrix, the target reflection signal frequency band in the adaptive enhancement feature map is compensated in the frequency domain and enhanced in the spatial domain to generate an optimized adaptive enhancement feature map. The optimization process includes frequency domain noise suppression and spatial domain detail enhancement.

[0170] Here's a specific example:

[0171] During an underwater archaeological mission, an underwater robot first extracted the harmonic interference frequency band of suspended debris scattering from the reflected light signal and generated dynamic suppression weights for the suspended debris scattering noise. Next, it analyzed the water transmission attenuation spectrum, specifically the attenuation at different depths, to generate dynamic compensation weights for the water transmission attenuation. These two weights were then fused at multiple scales using the frequency-domain and spatial-domain optimization units within a dual-complementary dynamic convolutional optimization network to form a frequency-domain and spatial-domain optimization coefficient matrix. Finally, based on this matrix, the target reflected signal frequency band in the adaptive enhancement feature map was subjected to frequency-domain compensation and spatial-domain enhancement, improving the quality of the feature map and making the shipwreck's outline more clearly visible.

[0172] This scheme improves the image quality of the adaptive enhancement feature map through multi-scale fusion of dynamic suppression weights and dynamic compensation weights, effectively suppresses the interference of suspended matter scattering noise and water body transmission attenuation on the target reflection signal, and provides high-quality image data for underwater robot target detection and motion control.

[0173] In order to further improve the accuracy and stability of underwater robot motion control, in some embodiments, step 105: generating trajectory correction parameters of the end effector of the manipulator and disturbance compensation coefficients of the thruster in the underwater robot based on the optimized adaptive enhancement feature map includes:

[0174] Step 701: extracting target surface local deformation parameters and propeller disturbance harmonic frequency distribution data from the optimized adaptive enhancement feature map.

[0175] In step 701, the propeller disturbance harmonic frequency distribution data refers to the harmonic frequency distribution characteristics of the propeller disturbance in the frequency domain, which is used to analyze the impact of the propeller operation on the suspended object scattering noise.

[0176] In this embodiment, the target surface local deformation parameters and propeller disturbance harmonic frequency distribution data are extracted from the optimized adaptive enhancement feature map. The target surface local deformation parameters are extracted based on the geometric curvature distribution data in the feature map, and the propeller disturbance harmonic frequency distribution data is extracted based on the noise spectrum characteristics in the feature map.

[0177] Step 702: Generate trajectory correction parameters for the end effector of the manipulator according to the local deformation parameters of the target surface, and generate a propeller disturbance compensation coefficient based on the propeller disturbance harmonic frequency distribution data.

[0178] In an embodiment of the present application, trajectory correction parameters of the end effector of the robotic arm are generated based on the local deformation parameters of the target surface to ensure that the motion path of the robotic arm matches the geometric characteristics of the target surface; a propeller disturbance compensation coefficient is generated based on the propeller disturbance harmonic frequency distribution data to suppress the secondary scattering noise of suspended matter caused by the propeller operation.

[0179] Here's a specific example:

[0180] During an underwater archaeological mission, an underwater robot was tasked with extracting a fragile artifact from a shipwreck. First, the robot extracted the local deformation parameters of the target artifact's surface caused by long-term water erosion using an optimized adaptive enhancement feature map. These parameters helped the robot understand subtle surface irregularities and the distribution of vulnerable areas. Simultaneously, the propeller's perturbation harmonic frequency distribution data was extracted to analyze the potential impact of the robot's motion on the surrounding environment. Next, trajectory correction parameters for the robot's end effector were generated based on these deformation parameters, enabling the robot to avoid vulnerable areas on the artifact's surface and precisely adjust the grasping path. Furthermore, a disturbance compensation coefficient was designed based on the propeller's perturbation harmonic frequency distribution data to reduce the impact of propeller vibration on the robot's positioning accuracy, ensuring stability as it approached the artifact. Ultimately, the robot successfully and safely extracted the artifact in a complex underwater environment, avoiding damage caused by improper operation or external disturbances. This demonstrates the advantages of this solution in high-precision and high-stability operations.

[0181] This scheme generates robot arm trajectory correction parameters and thruster disturbance compensation coefficients by extracting the local deformation parameters of the target surface and the harmonic frequency distribution data of the thruster disturbance from the optimized adaptive enhancement feature map, thereby improving the accuracy and stability of the underwater robot motion control, ensuring that the robot arm accurately operates the target and suppressing the suspended object scattering noise caused by the thruster disturbance.

[0182] Figure 2 A schematic diagram of the structure of a dual complementary dynamic convolution optimization system for underwater robot image processing provided in an embodiment of the present application is shown as follows: Figure 2 As shown, the system includes:

[0183] The receiving module 21 is used to receive the reflected light signal and the suspended matter concentration gradient distribution data when the polarized laser light source in the underwater robot excites a multi-band orthogonal polarized light sequence. The reflected light signal includes the suspended matter scattering polarization characteristics associated with the motion trajectory of the underwater robot, the polarization modulation characteristics of the reflected light on the surface of the working target, and the water body transmission attenuation spectrum characteristics.

[0184] The determination module 22 is configured to determine the frequency band of the main signal reflected by the working target according to the polarization modulation characteristics of the light reflected from the working target surface, and determine the frequency band of the suspended object scattered harmonic interference according to the suspended object scattered polarization characteristics.

[0185] The input module 23 is used to input the frequency band of the main signal reflected by the operation target into a preset dual complementary dynamic convolution optimization network, and output an adaptive enhanced feature map linked to the spatial coordinates of the operation target.

[0186] The optimization module 24 is configured to optimize the adaptive enhancement feature map by combining the suspended matter scattering harmonic interference frequency band and the water body transmission attenuation spectrum characteristics to obtain an optimized adaptive enhancement feature map.

[0187] The generation module 25 is used to generate the trajectory correction parameters of the end effector of the manipulator and the disturbance compensation coefficient of the propeller in the underwater robot based on the optimized adaptive enhancement feature map, so as to control the motion trajectory of the underwater robot.

[0188] Figure 2 The dual complementary dynamic convolution optimization system for underwater robot image processing can be executed Figure 1 The implementation principles and technical effects of the dual complementary dynamic convolution optimization method for underwater robot image processing described in the illustrated embodiment are not further elaborated. The specific manner in which each module and unit performs operations in the dual complementary dynamic convolution optimization system for underwater robot image processing in the above embodiment has been described in detail in the relevant embodiments of the method and will not be elaborated on here.

[0189] In one possible design, Figure 2 The dual complementary dynamic convolution optimization system for underwater robot image processing in the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32 .

[0190] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0191] The processing component 32 is as follows Figure 1 The embodiment provides a dual complementary dynamic convolution optimization method for underwater robot image processing.

[0192] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0193] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0194] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0195] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0196] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0197] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0198] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment provides a dual complementary dynamic convolution optimization method for underwater robot image processing.

[0199] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0200] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0201] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0202] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. 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 application.

Claims

1. A dual complementary dynamic convolution optimization method for underwater robot image processing, characterized by: include: In the case where a polarized laser light source in an underwater robot excites a multi-band orthogonal polarized light sequence, a reflected light signal and suspended matter concentration gradient distribution data are received, wherein the reflected light signal includes suspended matter scattering polarization characteristics associated with the underwater robot's motion trajectory, polarization modulation characteristics of light reflected from the surface of the operating target, and water body transmission attenuation spectrum characteristics; Determine the frequency band of the main signal reflected by the working target according to the polarization modulation characteristics of the reflected light on the surface of the working target, and determine the frequency band of the suspended object scattered harmonic interference according to the polarization characteristics of the suspended object scattered light; Input the frequency band of the main signal reflected by the operation target into a preset dual complementary dynamic convolution optimization network, and output an adaptive enhanced feature map linked to the spatial coordinates of the operation target; The adaptive enhancement feature map is optimized based on the suspended matter scattering harmonic interference frequency band and the water body transmission attenuation spectrum characteristics to obtain an optimized adaptive enhancement feature map; Based on the optimized adaptive enhancement feature map, the trajectory correction parameters of the end effector of the manipulator arm and the disturbance compensation coefficient of the propeller in the underwater robot are generated to control the motion trajectory of the underwater robot.

2. The method according to claim 1, characterized in that The method of inputting the frequency band of the main signal reflected by the operation target into a preset dual complementary dynamic convolution optimization network and outputting an adaptive enhanced feature map linked to the spatial coordinates of the operation target includes: Obtaining a layered scattering intensity weight and a band modulation phase parameter of the polarized laser light source, wherein the layered scattering intensity weight is generated by a preset multi-scale scattering model according to the suspended matter concentration gradient distribution data; generating a set of multi-scale scattering compensation coefficients based on the layered scattering intensity weights and the band modulation phase parameters; extracting time-varying polarization modulation features from the motion trajectory data fed back in real time by the underwater robot; The time-varying polarization modulation feature and the multi-scale scattering compensation coefficient set are cross-fused in the frequency domain and the spatial domain by a dynamic weight allocation unit in the dual complementary dynamic convolution optimization network to generate a fused feature set; In the dual complementary dynamic convolution optimization network, iteratively optimizing the fusion feature set; Constructing a three-dimensional geometric constraint field of the target contour based on the main signal frequency band reflected by the operation target and the motion trajectory data; Through the feature projection unit in the dual complementary dynamic convolution optimization network, the iteratively optimized fusion feature set is input into the three-dimensional geometric constraint field, and an adaptive enhanced feature map linked to the spatial coordinates of the operation target is output.

3. The method according to claim 2, characterized in that The feature projection unit in the dual complementary dynamic convolution optimization network inputs the iteratively optimized fusion feature set into the three-dimensional geometric constraint field, and outputs an adaptive enhanced feature map linked to the spatial coordinates of the operation target, including: generating an attenuation gradient distribution of a frequency band of a main signal reflected by the operation target; The attenuation gradient distribution is coupled with the local deformation parameters of the target surface in a multi-resolution space through a spatial topology adaptation unit in the three-dimensional geometric constraint field to obtain a coupling result; In the feature projection unit, based on the coupling result, dynamically correcting the spatial projection path of the iteratively optimized fusion feature set; The modified spatial projection path is associated with the propeller disturbance parameters of the underwater robot in real time through a geometric light field synchronization unit in the three-dimensional geometric constraint field to generate a projection path optimization sequence; Based on the projection path optimization sequence, the iteratively optimized fusion feature set is subjected to multi-channel feature reorganization according to the geometric constraint relationship of the spatial coordinates in the work target, and an adaptive enhanced feature map is generated that is synchronously linked to the light intensity distribution gradient of the polarization modulation characteristics of the reflected light on the target surface and the acceleration of the robot motion trajectory.

4. The method according to claim 3, characterized in that The dynamically correcting the spatial projection path of the iteratively optimized fusion feature set based on the coupling result includes: Based on the coupling result, cross-scale fusion is performed on the mapping weights between the local deformation parameters of the target surface and the attenuation gradient distribution at different levels of resolution to generate a multi-resolution deformation attenuation correlation tensor; According to the dynamic deformation correction coefficient of the three-dimensional geometric constraint field, at the global resolution level, the global projection path direction of the iteratively optimized fusion feature set in the three-dimensional geometric constraint field is corrected by using the displacement gradient direction of the contact surface between the thyroid margin and the carotid artery in the multi-resolution deformation attenuation correlation tensor; At the local resolution level, adjusting the local projection path weight of the iteratively optimized fusion feature set in the three-dimensional geometric constraint field by the spatial overlap between the microcalcification cluster density distribution in the multi-resolution deformation attenuated correlation tensor and the metabolomics signal hotspot area; Based on the superposition result between the global projection path direction and the local projection path weight, the spatial projection path of the iteratively optimized fusion feature is dynamically corrected.

5. The method according to claim 2, characterized in that The method of performing frequency-domain and spatial-domain cross-fusion of the time-varying polarization modulation feature and the multi-scale scattering compensation coefficient set by the dynamic weight allocation unit in the dual complementary dynamic convolution optimization network to generate a fused feature set includes: Performing multi-band decomposition on the time-varying polarization modulation characteristics according to the spatial distribution direction of the underwater robot's motion trajectory to obtain a set of frequency domain orthogonal basis functions; The frequency domain orthogonal basis function set and the multi-scale scattering compensation coefficient set are matched in spatial domain convolution kernel scale through the spatial domain feature mapping channel in the dynamic weight allocation unit. During the matching process, the coupling ratio of the frequency domain basis function weights of the frequency domain orthogonal basis function set and the spatial domain convolution kernel scale is dynamically adjusted according to the suspended matter concentration gradient distribution data to generate a frequency domain and spatial domain cross fusion coefficient matrix; According to the obstacle distance gradient information in the obstacle avoidance sensor data of the underwater robot, the cross fusion coefficient matrix is ​​dynamically sparsely processed to generate a fusion feature set.

6. The method according to claim 1, characterized in that The adaptive enhancement feature map is optimized by combining the suspended matter scattering harmonic interference frequency band and the water body transmission attenuation spectrum characteristics to obtain the optimized adaptive enhancement feature map, including: generating a dynamic suppression weight of suspended object scattering noise based on a set of harmonic side lobes in the suspended object scattering harmonic interference frequency band; generating a dynamic compensation weight for water body transmission attenuation according to the depth attenuation gradient distribution data in the water body transmission attenuation spectrum characteristics; The dynamic suppression weight of the suspended matter scattering noise and the dynamic compensation weight of the water body transmission attenuation are multi-scale fused through the frequency domain and space domain optimization unit in the dual complementary dynamic convolution optimization network to generate a frequency domain and space domain optimization coefficient matrix; Based on the frequency domain and spatial domain optimization coefficient matrix, frequency domain compensation and spatial domain enhancement are performed on the target reflection signal frequency band in the adaptive enhancement feature map to generate an optimized adaptive enhancement feature map.

7. The method according to claim 1, characterized in that The generating of trajectory correction parameters of the end effector of the manipulator and disturbance compensation coefficients of the propeller in the underwater robot based on the optimized adaptive enhancement feature map includes: Extracting target surface local deformation parameters and propeller disturbance harmonic frequency distribution data from the optimized adaptive enhanced feature map; According to the local deformation parameters of the target surface, a trajectory correction parameter of the end effector of the manipulator is generated, and based on the propeller disturbance harmonic frequency distribution data, a propeller disturbance compensation coefficient is generated.

8. A dual complementary dynamic convolution optimization system for underwater robot image processing, characterized by: include: A receiving module is configured to receive reflected light signals and suspended matter concentration gradient distribution data when a polarized laser light source excites a multi-band orthogonal polarized light sequence in the underwater robot. The reflected light signals include suspended matter scattering polarization characteristics associated with the underwater robot's motion trajectory, polarization modulation characteristics of light reflected from the surface of the operating target, and water body transmission attenuation spectrum characteristics. A determination module is configured to determine a frequency band of a main signal reflected by the working target according to the polarization modulation characteristics of the light reflected from the surface of the working target, and to determine a frequency band of a harmonic interference wave scattered by the suspended matter according to the polarization characteristics of the suspended matter scattering; An input module, configured to input the frequency band of the main signal reflected by the operation target into a preset dual complementary dynamic convolution optimization network, and output an adaptive enhanced feature map linked to the spatial coordinates of the operation target; an optimization module, configured to optimize the adaptive enhancement feature map by combining the suspended matter scattering harmonic interference frequency band and the water body transmission attenuation spectrum characteristics to obtain an optimized adaptive enhancement feature map; A generation module is used to generate trajectory correction parameters of the end effector of the manipulator arm and disturbance compensation coefficients of the propeller in the underwater robot based on the optimized adaptive enhancement feature map, so as to control the motion trajectory of the underwater robot.

9. A computing device, characterized in that It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the dual complementary dynamic convolution optimization method for underwater robot image processing as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, the dual complementary dynamic convolution optimization method for underwater robot image processing according to any one of claims 1 to 7 is implemented.

Images