Rain and fog weather instrument character recognition method and system fused with large model prior knowledge

Abstract

The invention provides a rain and fog weather instrument character recognition method and system fused with large model prior knowledge, and the method comprises the steps: collecting the hydrophobic characteristic data of the surface of an anticorrosive coating and the dynamic parameters of an environment when a navigation instrument is in a rain and fog environment, and generating an optical compensation strategy through the analysis of a large language model, and dynamically adjusting the wavelength of the light source, the polarization angle and the transmittance of the optical filter to form an interference fringe spectrum representing the characteristics of the liquid drops and the salt crystals. Salt crystal scattering noise is suppressed and liquid drop refraction distortion is corrected through double-branch processing, and an identification result is optimized by combining a space correlation technology and a character stroke continuity prior rule. According to the invention, the anti-interference capability and accuracy of navigation instrument character recognition in a rain and fog environment are improved.

Description

Technical Field

[0001] This application relates to the technical fields of computer vision and optical detection, and particularly to a method and system for identifying instrument characters in rainy and foggy weather conditions that incorporates prior knowledge of large models. Background Art

[0002] Navigation instruments are susceptible to interference from droplet attachment and salt crystal deposition in harsh environments such as salt spray, rain, and fog, resulting in refractive distortion and scattering noise in the instrument characters, seriously affecting the accuracy of visual recognition. A method that can dynamically adapt to environmental changes, suppress optical interference, and accurately identify characters is needed to ensure the safety and reliability of navigation operations.

[0003] One current solution to this problem is an identification method based on multi-spectral imaging and convolutional neural networks. By using multi-band light sources to collect instrument images under different environmental conditions, neural networks are used to suppress noise and distortion in the images, and finally, instrument information is extracted through a character segmentation algorithm.

[0004] This solution relies on fixed optical parameter configurations, resulting in problems such as residual scattering noise and insufficient correction of refractive distortion under extreme rainy and foggy conditions. At the same time, neural networks lack the utilization of prior knowledge of the instrument character structure, and have poor recognition robustness when strokes are broken or adhered. Summary of the Invention

[0005] This application provides a method and system for identifying instrument characters in rainy and foggy weather conditions that incorporates prior knowledge of large models, to solve the problems of low anti-interference ability and poor accuracy in identifying instrument characters in rainy and foggy environments in the prior art.

[0006] In a first aspect, this application provides a method for identifying instrument characters in rainy and foggy weather conditions that incorporates prior knowledge of large models, including: When the navigation instrument is in a rainy and foggy environment, obtain the surface hydrophobic property data and environmental dynamic parameters of the anti-corrosion coating of the navigation instrument; Perform correlation analysis on the surface hydrophobic property data and the environmental dynamic parameters through a pre-trained large language model to generate an optical compensation strategy; According to the optical compensation strategy, generate an interference fringe pattern corresponding to the surface of the anti-corrosion coating of the navigation instrument; Perform double-branch processing on the interference fringe pattern to generate a noise suppression pattern and a distortion correction pattern respectively; Perform spatial correlation on the noise suppression pattern and the distortion correction pattern to obtain a spatial correlation pattern, and iteratively optimize the spatial correlation pattern in combination with the prior rule of character stroke continuity in the preset large model prior knowledge base to generate an instrument character recognition result.

[0007] Optionally, iteratively optimize the spatial association graph by combining the character stroke continuity prior rule in the preset large model prior knowledge base to generate the instrument character recognition result, including: Extract candidate line segments related to the stroke direction of the instrument character from the spatial association graph; Match the candidate line segments with the character stroke continuity prior rule to generate a set of candidate line segments; Iteratively optimize the spatial association graph according to the set of candidate line segments to generate an optimized spatial association graph; Extract candidate character boundaries that match the predefined character structure template in the character stroke continuity prior rule from the optimized spatial association graph to generate the instrument character recognition result.

[0008] Optionally, the matching of the candidate line segments with the character stroke continuity prior rule to generate a set of candidate line segments includes: Divide the line segment direction type set according to the extension direction and length ratio of the candidate line segments in the spatial association graph; Extract the stroke turning angle range corresponding to the line segment direction type set from the character stroke continuity prior rule; Group the candidate line segments into candidate line segment clusters according to the corresponding extension direction; Compare the head and tail endpoint spacing of adjacent line segment pairs in the candidate line segment cluster with the preset spacing tolerance, where the head and tail endpoint spacing is the spacing between the head endpoint of one candidate line segment and the tail endpoint of another candidate line segment in the adjacent line segment pair; For each adjacent line segment pair with a head and tail endpoint spacing less than the spacing tolerance, calculate the angle fluctuation amplitude of the adjacent line segment pair based on the stroke turning angle range; Generate a set of candidate line segments based on the angle fluctuation amplitude.

[0009] Optionally, the generating a set of candidate line segments based on the angle fluctuation amplitude includes: Extract the allowable threshold of curvature change corresponding to the line segment direction type set from the character stroke continuity prior rule; Eliminate adjacent line segment pairs with an angle fluctuation amplitude exceeding the allowable threshold of curvature change within the candidate line segment cluster to generate a target line segment cluster; Taking the center point of the target line segment cluster as a reference, extend virtual connection lines along the curvature change direction, and judge whether the virtual connection lines are consistent with the stroke extension trend of the character structure in the character stroke continuity prior rule; When the judgment is consistent, dynamically adjust the angle threshold of adjacent line segments within the target line segment cluster; Connect and integrate adjacent line segment pairs that meet the dynamically adjusted included angle threshold to generate a candidate line segment set.

[0010] Optionally, the iteratively optimizing the spatial association map according to the candidate line segment set to generate an optimized spatial association map includes: Dividing a pixel confidence enhancement region associated with the overlapping probability of character strokes in the spatial association map according to the spatial distribution relationship between the endpoints of the line segments in the candidate line segment set; Based on the stroke connection topological constraint relationship in the prior rule of character stroke continuity, performing bidirectional weighted diffusion on the pixel confidence enhancement region to construct a confidence distribution network covering the complete character stroke contour; According to the confidence distribution network and the candidate line segment set, dynamically enhancing or attenuating the confidence level of pixel points in the spatial association map to generate an optimized spatial association map.

[0011] Optionally, the associatively analyzing the surface hydrophobic property data and the environmental dynamic parameters by a pre-trained large language model to generate an optical compensation strategy includes: Binding the coating contact angle change rate in the surface hydrophobic property data and the salt fog concentration gradient in the environmental dynamic parameters in a time series to generate a coupling feature vector of the environment and the material; Retrieving and matching historical optical compensation cases in the large language model based on the coupling feature vector, and extracting a case set whose similarity to the current coupling feature vector reaches a preset similarity threshold from the historical optical compensation cases; According to the case set, calculating the predicted value of the droplet distribution density and the predicted value of the salt crystal deposition thickness on the surface of the anti-corrosion coating of the marine instrument under the current environmental conditions; Generating corresponding light source wavelength adjustment suggestion information and polarization angle adjustment suggestion information according to the predicted value of the droplet distribution density; Generating corresponding filter transmittance curve adjustment suggestion information according to the predicted value of the salt crystal deposition thickness; Integrating the light source wavelength adjustment suggestion information, the polarization angle adjustment suggestion information, and the filter transmittance curve adjustment suggestion information to generate an optical compensation strategy.

[0012] Optionally, the generating an interference fringe pattern corresponding to the surface of the anti-corrosion coating of the marine instrument according to the optical compensation strategy includes: Adjusting the wavelength of the incident beam of the illumination light source based on the light source wavelength adjustment suggestion information in the optical compensation strategy so that the incident beam excites a scattered characteristic beam representing the droplet morphology on the surface of the anti-corrosion coating; Adjust the polarization angle of the circularly polarized light source based on the polarization angle adjustment recommendation information in the optical compensation strategy to suppress the stray reflected light caused by the surface salt crystal deposition of the anti-corrosion coating of the marine instrument; Adjust the filter transmittance curve of the filter array based on the filter transmittance curve adjustment recommendation information in the optical compensation strategy to enhance the interference contrast between the scattered characteristic light beam and the refracted characteristic light beam; Generate an interference light intensity distribution through the interaction of the scattered characteristic light beam and the refracted characteristic light beam on the imaging plane, and generate an interference fringe pattern according to the interference light intensity distribution.

[0013] In a second aspect, the present application provides a rain and fog weather instrument character recognition system integrating large model prior knowledge, including: An acquisition module, configured to acquire the surface hydrophobic property data and environmental dynamic parameters of the anti-corrosion coating of the marine instrument when the marine instrument is in a rain and fog environment; An association analysis module, configured to perform association analysis on the surface hydrophobic property data and the environmental dynamic parameters through a pre-trained large language model to generate an optical compensation strategy; A first generation module, configured to generate an interference fringe pattern corresponding to the surface of the anti-corrosion coating of the marine instrument according to the optical compensation strategy; A second generation module, configured to perform double-branch processing on the interference fringe pattern to respectively generate a noise suppression pattern and a distortion correction pattern; A spatial association module, configured to perform spatial association on the noise suppression pattern and the distortion correction pattern to obtain a spatial association pattern, and iteratively optimize the spatial association pattern in combination with the character stroke continuity prior rule in the preset large model prior knowledge base to generate an instrument character recognition result.

[0014] In a third aspect, the present application provides a computing device, including a processor and a memory, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the rain and fog weather instrument character recognition method according to any one of the first aspects.

[0015] In a fourth aspect, the present application provides a computer storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the rain and fog weather instrument character recognition method according to any one of the first aspects is implemented.

[0016] In this application, a method for identifying instrument characters in rainy and foggy weather conditions that incorporates the prior knowledge of large models is provided. The method includes: when the navigation instrument is in a rainy and foggy environment, obtaining the surface hydrophobic property data and environmental dynamic parameters of the anti-corrosion coating of the navigation instrument; performing correlation analysis on the surface hydrophobic property data and the environmental dynamic parameters through a pre-trained large language model to generate an optical compensation strategy; generating an interference fringe pattern corresponding to the surface of the anti-corrosion coating of the navigation instrument according to the optical compensation strategy; performing a two-branch process on the interference fringe pattern to respectively generate a noise suppression pattern and a distortion correction pattern; performing spatial correlation on the noise suppression pattern and the distortion correction pattern to obtain a spatial correlation pattern, and iteratively optimizing the spatial correlation pattern in combination with the prior rule of character stroke continuity in a preset large model prior knowledge base to generate an instrument character recognition result.

[0017] The technical solution provided by this application has the following beneficial effects: This application can perceive the surface state of the instrument and environmental changes in real time, providing a data basis for optical compensation. It can dynamically adapt to different rainy and foggy conditions and optimize the configuration of light source parameters. At the same time, it can capture the characteristics of salt crystal scattering and droplet refraction to form anti-interference optical representations. It can respectively eliminate particle noise and geometric deformation. It can fuse multi-modal features and use prior knowledge to repair the stroke structure.

[0018] Furthermore, this application also extracts candidate line segments from the spatial correlation pattern, combines the character stroke continuity rule to screen the effective line segment set, and after iterative optimization, matches the predefined character template, and finally outputs the recognition result.

[0019] Moreover, it can effectively solve the problems of character stroke breakage / adhesion in rainy and foggy environments, and improve the accuracy of character structure restoration in complex interference scenarios through the optimization process guided by prior knowledge.

[0020] These aspects or other aspects of this application will be more clearly understood in the following description of the embodiments. Description of the Drawings

[0021] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of this application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0022] Figure 1 It is a flowchart of a method for identifying instrument characters in rainy and foggy weather conditions that incorporates the prior knowledge of large models provided by an embodiment of this application; Figure 2Schematic structural diagram of a rain and fog weather instrument character recognition system integrating prior knowledge of large models provided by an embodiment of the present application; Figure 3 Schematic structural diagram of a computing device provided by an embodiment of the present application. Detailed implementation manners

[0023] In order to enable those skilled in the art to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application.

[0024] In some processes described in the specification, claims and the above-mentioned drawings of the present application, a plurality of operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear herein or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. 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 such as "first" and "second" in this article are used to distinguish different messages, devices, modules, etc., and do not represent a sequence, nor do they limit that "first" and "second" are of different types.

[0025] Researchers have found that due to salt crystal deposition and droplet attachment in the rain and fog environment, traditional visual recognition methods for navigation instruments are seriously interfered, and it is difficult for the existing technologies to dynamically adapt to environmental changes and effectively restore damaged character structures. Based on this, the embodiments of the present application provide a rain and fog weather instrument character recognition method integrating prior knowledge of large models. This method dynamically generates an optical compensation strategy by using a large language model through real-time perception of the hydrophobic characteristics of the instrument surface and environmental parameters, combines dual-branch processing to separate noise and distortion features, and finally realizes anti-interference recognition through iterative optimization guided by prior knowledge. The technical solution of the present application is applicable to instrument automatic recognition scenarios under harsh weather conditions such as navigation and aviation.

[0026] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0027] Figure 1 Flowchart of a rain and fog weather instrument character recognition method integrating prior knowledge of large models provided by an embodiment of the present application, as Figure 1 shown, the method includes: Step 101: When the marine instrument is in a rain and fog environment, obtain the surface hydrophobic property data of the anti-corrosion coating of the marine instrument and the environmental dynamic parameters.

[0028] In this step, the surface hydrophobic property data represents the ability of the anti-corrosion coating to repel droplets and is used to predict the attachment morphology of rain and fog. The environmental dynamic parameters represent the key external factors such as the salt fog concentration, humidity, and temperature monitored in real time, which affect the optical imaging quality.

[0029] In the embodiment of the present application, the contact angle data of the droplets on the surface of the instrument anti-corrosion coating is collected by a contact angle measuring instrument, and the salt fog concentration and humidity data are obtained synchronously by using an environmental sensor. After aligning the two types of data according to the time stamp, they are input into the system. The contact angle data reflects the current hydrophobic performance of the coating, and the environmental parameters are used to evaluate the interference intensity.

[0030] For example, taking a marine engine instrument that displays "Rotation speed 1500 RPM" as an example, the contact angle of the coating is measured to be 120 degrees (indicating high hydrophobicity), and at the same time, a high-level salt fog concentration and a saturated humidity are detected. These data indicate that serious droplet attachment and salt crystal scattering will occur.

[0031] Step 102: Perform correlation analysis on the surface hydrophobic property data and the environmental dynamic parameters through a pre-trained large language model to generate an optical compensation strategy.

[0032] In this step, the optical compensation strategy includes adjustment schemes such as the light source wavelength, polarization angle, and filter parameters, which are used to offset environmental interference.

[0033] In the embodiment of the present application, the large language model analyzes the correlation law between the contact angle data and the salt fog concentration: when the contact angle is greater than the threshold and the salt fog concentration is high, it outputs a short-wavelength light source to enhance the contrast of the droplet edge; adjusts the polarization angle of the polarizer according to the humidity value to suppress the salt crystal reflection. The model generates a parameter combination through historical case matching.

[0034] For example, for the aforementioned 120-degree contact angle and high salt fog environment, the model selects 450nm blue light (short wavelength) to highlight the droplet contour, sets the polarization angle of the polarizer at 60 degrees to filter out the salt crystal stray light, and forms a targeted optical compensation scheme.

[0035] Step 103: Generate an interference fringe pattern corresponding to the surface of the anti-corrosion coating of the marine instrument according to the optical compensation strategy.

[0036] In this step, the interference fringe pattern represents a two-dimensional grayscale image formed by the interference of scattered light (salt crystal) and refracted light (droplet), which contains intensity and phase information.

[0037] In the embodiments of the present application, a multi-spectral light source and a polarizer are configured according to a compensation strategy. After irradiating the surface of the instrument, the reflected light is separated into characteristic bands by a filter array. An imaging sensor collects the interference light intensity distribution. Radial stripes are formed by the scattering of salt crystals, and concentric circles are generated by the refraction of droplets, and a composite map is synthesized.

[0038] For example, when irradiating a "1500RPM" instrument with 450nm blue light and imaging after polarization filtering. The strokes of the digit "1" produce parallel stripes with a spacing of 2mm due to droplets (determined by the wavelength and droplet curvature), and curved stripes appear at the arcs of the letter "R".

[0039] Step 104: Perform a two-branch process on the interference fringe map to generate a noise suppression map and a distortion correction map respectively.

[0040] In this step, the noise suppression map represents an image after eliminating the salt crystal scattering noise, retaining the main body of the characters. The distortion correction map represents an image that corrects the geometric deformation caused by droplet refraction.

[0041] In the embodiments of the present application, the first branch uses a polarization difference algorithm to suppress salt crystal noise by comparing the image differences in different polarization directions; the second branch performs reverse calculation based on the refraction optical path to restore the distorted strokes to their original shapes. The two branches process the same interference map in parallel.

[0042] For example, in the processing of "1500RPM", the first branch eliminates the salt crystal noise around the letter "P", and the second branch corrects the bending deformation of the digit "5" caused by droplets, restoring its standard semi-circular structure.

[0043] Step 105: Perform spatial correlation on the noise suppression map and the distortion correction map to obtain a spatial correlation map, and iteratively optimize the spatial correlation map in combination with the prior rules of character stroke continuity in the preset large model prior knowledge base to generate the instrument character recognition result.

[0044] In this step, the spatial correlation map represents a three-dimensional data volume that fuses the features of the two types of maps, including coordinate, intensity, and confidence information. The prior rules of character stroke continuity represent the stroke connection probability library provided by the large model, such as the digit "1" should be a continuous straight line. The instrument character recognition result refers to the readable character information finally output after suppressing environmental interference, feature fusion, and prior knowledge optimization. Specifically, it accurately restores the real content displayed on the navigation instrument in rainy and foggy environments (such as "rotation speed 1500RPM"), where the stroke continuity and geometric structure of each character are consistent with the actual instrument display, and the artifacts caused by salt crystal scattering and the deformation caused by droplet refraction are eliminated, ensuring the integrity and recognizability of numbers, letters, and unit symbols, and meeting the precise interpretation requirements of instrument data in navigation operations.

[0045] In the embodiments of the present application, two types of atlases are aligned according to pixel coordinates, and confidence weights are calculated for the overlapping regions. The prior rules guide the optimization: when a break in the stroke of the digit "1" is detected, the missing pixels are filled according to the line continuity; the adhered strokes are segmented according to the character template.

[0046] For example, the "P" in "RPM" is interrupted by noise. The prior rules recognize that it should contain a closed ring structure and automatically connect the broken edges; the middle part of the digit "0" is blurred by droplets, and the edge confidence is enhanced according to the circular template.

[0047] This method dynamically senses the environment and the coating state, generates a targeted optical compensation scheme, uses interference imaging to separate the interference characteristics of salt crystals and droplets, suppresses noise and corrects deformation through double-branch processing, and finally achieves accurate recognition by combining the prior knowledge of the character structure.

[0048] To solve the problems of stroke breakage and deformation in instrument character recognition in rainy and foggy environments and further improve the recognition accuracy, in some embodiments, step 105: iteratively optimizing the spatial association atlas by combining the prior rules of character stroke continuity in the preset large model prior knowledge base to generate the instrument character recognition result, including: Step 201: Extract candidate line segments related to the stroke direction of the instrument characters from the spatial association atlas.

[0049] In step 201, the candidate line segments refer to a set of line segments suspected of character strokes extracted from the spatial association atlas. Each line segment contains start point coordinates, end point coordinates, and direction angle information, and is used to preliminarily represent the character structure.

[0050] In the embodiments of the present application, the connected regions in the spatial association atlas are extracted through an edge detection algorithm, the principal axis directions of each region are calculated, and a set of line segments with consistent directions is generated, filtering out the interference line segments with too short length or disordered directions.

[0051] Step 202: Match the candidate line segments with the prior rules of character stroke continuity to generate a set of candidate line segments.

[0052] In step 202, the set of candidate line segments refers to the line segment combinations retained after being screened by the prior rules of character stroke continuity. These line segments conform to the stroke feature rules of the predefined character structure in terms of direction continuity, turning angle, and curvature change. Specifically, it includes two types of valid line segments: one is the originally complete character stroke segments in the spatial association atlas, and the other is the broken stroke segments filled by the prior rules. This set serves as the basic input for subsequent iterative optimization to ensure the structural integrity and geometric accuracy of the finally recognized characters. For example, when recognizing the oil pressure gauge "25.5MPa", this set includes the complete S-shaped stroke segment of the digit "5" filled by the rules and the straight stroke segments of the letters in the unit "MPa".

[0053] In the embodiment of the present application, after grouping the candidate line segments by direction, they are matched with the character templates in the rule library, and the line segment combinations with turning angles within a preset range and smooth curvature changes are screened out to form a candidate line segment set.

[0054] Step 203: According to the candidate line segment set, iteratively optimize the spatial association map to generate an optimized spatial association map.

[0055] In step 203, the optimized spatial association map refers to a feature map after stroke continuity correction, and its pixel confidence distribution more conforms to the real character structure.

[0056] In the embodiment of the present application, according to the connection relationship of each line segment in the candidate line segment set, the pixel confidence of the line segment break region in the original map is enhanced, and the pixel confidence of the line segment intersection conflict region is attenuated. After multiple iterations, an optimized map is output.

[0057] Step 204: Extract candidate character boundaries that match the predefined character structure templates in the character stroke continuity prior rule from the optimized spatial association map to generate an instrument character recognition result.

[0058] In step 204, the candidate character boundary refers to a closed contour line in the optimized map that satisfies the geometric features of the character template and is used for final recognition.

[0059] In the embodiment of the present application, the connected domain boundary is extracted on the optimized map, and its shape is matched with the character templates in the rule library, and the boundaries with a similarity reaching the threshold are selected as the recognition result.

[0060] The following is a specific example: Taking the marine hydraulic pressure gauge showing "hydraulic pressure 25.5 MPa" as an example, in an environment where the measured coating contact angle is 115 degrees and the salt spray concentration is at a high level, 470 nm blue-green light (calculated by the optical compensation strategy according to the relationship between droplet curvature and wavelength refractive index) is used for irradiation to generate interference fringes: The number "5" forms ripple fringes with a spacing of 1.8 mm due to droplet refraction (spacing = wavelength × droplet curvature radius / refractive index difference), and the slashes of the letter "M" present intermittent bright spots due to salt crystal scattering. In the dual-branch processing, the first branch eliminates the salt crystal noise on the right side of the "M", and the second branch corrects the S-shaped distortion caused by the droplet at the lower part of the number "5"; during the spatial correlation optimization, the prior rule identifies the characteristic of the double-curved line segment that the number "5" should have, complements the line segment broken at the wave trough according to the curvature continuity, and repairs the circular structure of the blurred "a" character in the unit "MPa" according to the standard font template, and finally outputs the complete and accurate recognition result of "25.5 MPa". Among them, the 1.8 mm fringe spacing is calculated through the optical formula, specifically: the selected wavelength of 470 nm is multiplied by the average droplet curvature radius of 3.8 mm (converted through the contact angle) and divided by the refractive index difference of 1.33 between oil and water.

[0061] In the embodiment of the present application, through the above steps, the structural repair and accurate recognition of broken and deformed characters under rain and fog interference are realized, and the reliability of instrument readings in harsh environments is improved.

[0062] To solve the problem of difficult matching caused by broken and deformed strokes of instrument characters in rain and fog environments and further improve the accuracy of character line segment screening, in some embodiments, step 202: The step of matching the candidate line segments with the prior rules of character stroke continuity to generate a set of candidate line segments includes: Step 301: Divide the set of line segment direction types according to the extension direction and length ratio of the candidate line segments in the spatial correlation map.

[0063] In step 301, the set of line segment direction types refers to the categories obtained by dividing the candidate line segments according to their main extension directions in the spatial correlation map, including four basic types: horizontal, vertical, left oblique, and right oblique. Each type of line segment has similar length ratio characteristics.

[0064] In the embodiment of the present application, by calculating the angles between each line segment and the coordinate axes and combining the ratio of the line segment length to the average stroke length of the character, the candidate line segments are classified into the preset set of direction types.

[0065] Step 302: Extract the stroke turning angle range corresponding to the set of line segment direction types from the prior rules of character stroke continuity.

[0066] In step 302, the stroke turning angle range refers to the maximum angle change threshold allowed when the character stroke changes between different direction types.

[0067] In the embodiment of the present application, the character type being currently processed (such as numbers or letters) is queried from the prior rule library, and the reasonable turning angle range of the character during conversion in each direction type is extracted.

[0068] Step 303: Group the candidate line segments into candidate line segment clusters according to their corresponding extension directions.

[0069] In step 303, a candidate line segment cluster refers to a grouping of line segments of the same direction type and adjacent in spatial position, and is used for local stroke continuity analysis.

[0070] In the embodiment of the present application, line segments with the same direction type are clustered according to their coordinate positions in the map to form a line segment combination that may belong to the same stroke in space.

[0071] Step 304: Compare the distance between the head and tail endpoints of adjacent line segment pairs in the candidate line segment cluster with a preset distance tolerance. The distance between the head and tail endpoints is the distance between the head endpoint of one candidate line segment and the tail endpoint of another candidate line segment in the adjacent line segment pair.

[0072] In step 304, the distance tolerance refers to the maximum endpoint distance threshold for determining whether two line segments may belong to the same stroke, and is dynamically adjusted according to the character size. Adjacent line segments refer to two or more candidate line segments with similar extension directions and the distance between the head and tail endpoints less than the preset tolerance in the spatial association map.

[0073] In the embodiment of the present application, the actual distance between the endpoints of adjacent line segments is calculated and compared with the preset tolerance value to screen out possible connected line segment pairs.

[0074] Step 305: For each adjacent line segment pair with the distance between the head and tail endpoints less than the distance tolerance, calculate the angle fluctuation amplitude of the adjacent line segment pair based on the stroke turning angle range.

[0075] In step 305, the angle fluctuation amplitude refers to the degree of deviation of the angle difference between the direction changes of adjacent line segments from the standard turning angle in the prior rules.

[0076] In the embodiment of the present application, for the line segment pairs screened by the distance, calculate the angle between their extension directions and compare it with the standard turning angle range in the rule library to obtain the fluctuation amplitude.

[0077] Step 306: Generate a candidate line segment set based on the angle fluctuation amplitude.

[0078] In the embodiment of the present application, line segment pairs with the fluctuation amplitude exceeding the threshold are excluded, and the qualified line segments are merged according to the spatial relationship to form a final candidate set.

[0079] The following is a specific example: Taking the identification of the digit "5" in the oil pressure gauge "25.5MPa" as an example, three broken line segments are extracted in the spatial association map: the first segment is in the right oblique 45-degree direction with a length of 3.2 mm (calculated from the interference fringe spacing of 1.8 mm, 1.8 mm×1.8≈3.2 mm), the second segment is in the horizontal direction with a length of 2.1 mm, and the third segment is in the left oblique 135-degree direction with a length of 3.0 mm. According to the prior rule of character stroke continuity, the standard turning angles of the digit "5" should be: turning from right oblique to horizontal at 120 degrees (±15 degrees), and turning from horizontal to left oblique at 135 degrees (±10 degrees). After grouping the three line segments by direction, the measured end-point spacings of adjacent line segments are 0.5 mm and 0.6 mm respectively (less than the preset tolerance of 1 mm). It is calculated that the included angle between the first segment and the second segment is 135 degrees (within the allowable range), and the included angle between the second segment and the third segment is 135 degrees (meeting the standard). Finally, the three line segments are combined into a complete S-shaped stroke. The 1 mm spacing tolerance is set according to the empirical value of 20% of the digit height of 5 mm to ensure the effective connection of broken strokes.

[0080] In the embodiment of the present application, through the above steps, accurate screening and recombination of broken strokes are achieved, effectively solving the recognition error problem caused by broken character strokes in rainy and foggy environments, and improving the character structure restoration ability in complex interference scenarios.

[0081] To solve the matching difficulty problem caused by broken and deformed instrument character strokes in rainy and foggy environments and further improve the accuracy of character line segment screening, in some embodiments, step 306: generating a candidate line segment set based on the included angle fluctuation range, including: Step 401: Extract the allowable threshold of curvature change corresponding to the line segment direction type set from the prior rule of character stroke continuity.

[0082] In step 401, the allowable threshold of curvature change refers to the maximum amount of curvature change allowed when the character stroke changes direction.

[0083] In the embodiment of the present application, according to the current character type and line segment direction type being processed, the corresponding curvature change threshold range is extracted from the prior rule library to judge the rationality of line segment turning.

[0084] Step 402: Remove the adjacent line segment pairs with the included angle fluctuation range exceeding the allowable threshold of curvature change within the candidate line segment cluster to generate a target line segment cluster.

[0085] In step 402, the target line segment cluster refers to the line segment grouping retained after being screened by the curvature change threshold, and the turning angles and curvature changes of its adjacent line segment pairs conform to the character structure characteristics.

[0086] In the embodiments of the present application, the curvature change amount of each pair of adjacent line segments in the candidate line segment cluster is calculated, the line segment pairs exceeding the allowable threshold are removed, and the line segment combinations with smooth turning are retained.

[0087] Step 403: Taking the center point of the target line segment cluster as a reference, extend a virtual connection line along the curvature change direction, and determine whether the virtual connection line is consistent with the stroke extension trend of the character structure in the prior rule of character stroke continuity.

[0088] In step 403, the curvature change direction is calculated from the geometric features of the candidate line segment cluster, and is specifically determined by the tangent direction change trend of adjacent line segments in the line segment cluster. The relationship with the allowable curvature change threshold: the actual fluctuation amplitude of the curvature change direction needs to be less than or equal to the allowable curvature change threshold, and the two together constitute the dynamic constraint condition for judging stroke continuity. The virtual connection line refers to a hypothetical connection line starting from the center point of the target line segment cluster and extending along the curvature change direction of the line segment cluster, used to simulate the extension path of an ideal character stroke. Its generation process is: first calculate the average curvature direction of each line segment in the target line segment cluster, and then extend a virtual reference line along this direction.

[0089] In the embodiments of the present application, the average curvature direction of each line segment in the target line segment cluster is calculated, and a virtual reference line is extended along this direction to match the standard character stroke extension trend in the rule library.

[0090] Step 404: When the judgment is consistent, dynamically adjust the included angle threshold between adjacent line segments in the target line segment cluster.

[0091] In step 404, dynamically adjusting the included angle threshold means adaptively relaxing or tightening the allowable range of the included angle between adjacent line segments according to the matching degree between the virtual connection line and the standard stroke trend.

[0092] In the embodiments of the present application, when the virtual connection line is highly consistent with the standard trend, the included angle threshold is appropriately relaxed to accommodate recoverable broken strokes; when the matching degree is low, the threshold is tightened to avoid incorrect connections.

[0093] Step 405: Connect and integrate the adjacent line segment pairs that meet the dynamically adjusted included angle threshold to generate a candidate line segment set.

[0094] In step 405, connecting and integrating means smoothly connecting the adjacent line segment pairs that meet the adjusted included angle threshold to form a complete stroke segment. The target line segment cluster refers to the line segment grouping after preliminary screening (removing the line segment pairs that do not meet the initial threshold), and the candidate line segment set is the line segment combination that is finally confirmed as a valid character stroke after the target line segment cluster is verified by the virtual connection line and dynamically threshold-adjusted. The candidate line segment set is a subset of the target line segment cluster and has a higher confidence level of stroke continuity.

[0095] In the embodiment of the present application, for the screened adjacent line segment pairs, smooth transition processing is performed at their endpoints to generate continuous stroke segments, which are then added to the candidate line segment set. Adjacent line segment pairs that do not meet the dynamically adjusted included angle threshold will remain separated and will not be connected and integrated, but they are still retained in the target line segment cluster as candidate objects for subsequent processing.

[0096] The following is a specific example: Taking the recognition of the digit "5" in the oil pressure gauge "25.5MPa" as an example, during the processing of the target line segment cluster, first, the allowable threshold for the curvature change of the digit "5" is extracted from the prior rules as no more than 40 degrees of rotation per millimeter (obtained by counting from the standard font library). The actual curvature changes of the first right-angled inclined line segment and the second horizontal line segment are measured to be 35 degrees / mm (calculated by dividing the included angle of 135 degrees of the line segment by the length of 3.8 mm at the turning point), and the curvature change between the second and the third line segments is 38 degrees / mm, both meeting the threshold requirements. Taking the center points of the three line segments as the reference to extend the virtual S-shaped connection line, the matching degree of its curvature change trend with the standard template of the digit "5" in the rule library reaches 92% (calculated by the curvature similarity algorithm). Therefore, the adjacent line segment included angle threshold is dynamically adjusted from ±15 degrees to ±18 degrees. Finally, the three line segments are smoothly connected at the turning points. Among them, a transition arc with a radius of 2.5 mm (radius = line segment length × curvature compensation coefficient 0.8) is used at the connection between the first and the second line segments to generate a complete and coherent candidate stroke for the digit "5".

[0097] In the embodiment of the present application, through the above steps, the intelligent screening and precise repair of broken strokes are realized, effectively solving the problem of difficult recognition caused by broken and deformed character strokes in rainy and foggy environments, and improving the character recognition accuracy in complex interference scenarios.

[0098] To solve the problem of difficult recognition caused by broken and deformed instrument character strokes in rainy and foggy environments and further improve the accuracy and robustness of character recognition, in some embodiments, step 203: iteratively optimizing the spatial association graph according to the candidate line segment set to generate an optimized spatial association graph includes: Step 501: According to the spatial distribution relationship between the endpoints of the line segments in the candidate line segment set, divide the pixel confidence enhancement regions associated with the overlapping probability of character strokes in the spatial association graph.

[0099] In step 501, the spatial distribution relationship between the endpoints of the line segments refers to the relative positions and connection trends of the endpoints of each line segment in the candidate line segment set, including features such as the distance between endpoints and the consistency of the extension direction. This relationship is obtained by calculating the actual distance and direction angle between adjacent line segment endpoints and is used to determine whether the broken line segments may belong to the same stroke. For example, in the recognition of the digit "5", the distance between the endpoints and the trend of direction change of the three line segments conform to the distribution law of the S-shaped stroke. The pixel confidence enhancement region refers to the pixel region in the spatial association map that overlaps or is adjacent to the spatial position of the line segments in the candidate line segment set, and these regions have a relatively high probability of the existence of character strokes.

[0100] In the embodiment of the present application, by calculating the spatial distance and direction consistency between the candidate line segment endpoints, the pixel regions that may belong to the same character stroke are marked in the spatial association map to form a confidence enhancement region.

[0101] Step 502: Based on the stroke connection topological constraint relationship in the prior rule of character stroke continuity, perform bidirectional weighted diffusion on the pixel confidence enhancement region to construct a confidence distribution network covering the complete character stroke contour.

[0102] In step 502, the stroke connection topological constraint relationship refers to the geometric connection rules of character strokes in the standard structure, including topological features such as the stroke turning angle and the curvature change trend. This relationship is extracted from the prior knowledge base of the large model and is used to guide the propagation direction of confidence in the network. For example, the constraint relationship of the digit "5" requires the stroke to turn continuously in an S shape and the curvature to change smoothly. The character stroke overlap probability refers to the possibility that a certain region in the spatial association map actually has a character stroke (rather than noise or distortion). The stroke direction of the instrument character refers to the directional feature of the strokes of the nautical instrument characters (such as geometric trends like straight lines and arcs). The complete character stroke contour refers to the optimized closed stroke boundary without breaks and distortions. The stroke direction is used to preliminarily screen candidate line segments; the overlap probability determines which line segment regions need to be strengthened; the complete contour is the final result of the collaborative optimization of the former two, and the three form a progressive correction chain from local features to global structure. The confidence distribution network refers to the pixel confidence propagation network established based on the stroke connection topological constraint relationship and is used to represent the complete contour of the character stroke.

[0103] In the embodiment of the present application, according to the connection relationship of the character strokes in the prior rule, starting from the central pixel of the confidence enhancement region, perform bidirectional weighted diffusion along the stroke extension direction, and gradually construct a confidence distribution network covering the entire character contour.

[0104] Step 503: According to the confidence distribution network and the candidate line segment set, dynamically enhance or attenuate the confidence level of the pixel points in the spatial association map to generate an optimized spatial association map.

[0105] In step 503, dynamic enhancement or attenuation refers to adjusting the confidence level of pixel points in the spatial association map according to the spatial relationship between the confidence distribution network and the candidate line segment set. Specifically: when the pixel points in the spatial association map meet the following two conditions simultaneously, confidence enhancement is performed: (1) being located on the high-weight path of the confidence distribution network, and (2) matching the geometric features of the line segments in the candidate line segment set; conversely, when the pixel points meet any of the following conditions, confidence attenuation is performed: (1) being in the low-weight area of the confidence distribution network, and (2) having a spatial position conflict with the candidate line segment set.

[0106] In the embodiment of the present application, for the pixel points on the high-weight path in the confidence distribution network and whose spatial positions match the line segments in the candidate line segment set, their confidence is enhanced; for the pixel points in the low-weight area or in conflict with the candidate line segments, their confidence is attenuated, and finally an optimized spatial association map is generated.

[0107] The following is a specific example: Taking the recognition of the digit "5" in the oil pressure gauge "25.5MPa" as an example, in the spatial association map, an S-shaped stroke area is divided as the confidence enhancement area according to the endpoint spacing (0.5mm and 0.6mm respectively) and the extension direction (right oblique 45 degrees to horizontal 0 degrees and then to left oblique 135 degrees) of the three line segments in the candidate line segment set. The width of this area is set to 1.2mm (determined by 40% of the average line segment length of 3.1mm). Based on the double-bending topological feature required by the prior rule of the digit "5", a two-way diffusion is carried out from the center of the area to both ends with a weight coefficient of 0.15 per pixel (calculated by the ratio of the stroke width to the length), and a complete S-shaped confidence network is constructed. The confidence of the pixel points at the two turning points (the area with a curvature radius of 2.5mm) of the line segments in the network is enhanced by 1.5 times, and the confidence of the pixel points in the middle blurred area (width about 0.8mm) caused by droplet interference is attenuated to 0.7 times, and finally an optimized map of the clear and coherent digit "5" is generated. Among them, the weight coefficient 0.15 = line segment width 1.2mm / (total line segment length 8.3mm × 10), and the 1.5-fold enhancement coefficient is calculated according to the matching degree of the curvature at the turning point with the standard value.

[0108] In the embodiment of the present application, through the above steps, the accurate repair and optimization of the broken and deformed strokes are realized, the interference problem in character recognition in rainy and foggy environments is effectively solved, and the accuracy and reliability of instrument character recognition are improved.

[0109] To solve the adaptability problem of the optical compensation strategy in the rain and fog environment and further improve the accuracy of instrument character recognition, in some embodiments, step 102: The correlation analysis of the surface hydrophobic property data and the environmental dynamic parameters by the pre-trained large language model to generate an optical compensation strategy includes: Step 601: Bind the coating contact angle change rate in the surface hydrophobic property data and the salt fog concentration gradient in the environmental dynamic parameters in a time series to generate a coupling feature vector of the environment and the material.

[0110] In step 601, the coating contact angle change rate refers to the change amount of the contact angle on the surface of the anti-corrosion coating per unit time, reflecting the dynamic change trend of the hydrophobic performance. This data is obtained by collecting the contact angle values at multiple positions on the surface of the instrument in real time by a contact angle measuring instrument and calculating the change slope over time, and is used to quantify the influence degree of rain and fog adhesion on the hydrophobicity of the coating. For example, in the oil pressure gauge scenario, the change rate of the contact angle measured from the initial 120 degrees to 115 degrees is 5 degrees / hour. The salt fog concentration gradient refers to the change rate of the salt fog particle concentration within a unit spatial range, characterizing the uneven distribution of salt fog deposition. This data is collected by multi-point salt fog sensors arranged around the instrument and obtained by calculating the spatial difference between the detection values of adjacent sensors. For example, in the navigation scenario, it is measured that the concentration above the instrument is 15% / cm higher than that on the side, forming a concentration gradient in the vertical direction. The coupling feature vector of the environment and the material refers to the multi-dimensional feature representation formed by the spatio-temporal correlation of the surface hydrophobic property and the environmental parameters, and is used to characterize the physical state of the material surface under specific environmental conditions.

[0111] In the embodiment of the present application, the change rate of the coating contact angle over time is synchronously aligned with the salt fog concentration change curve, and a multi-dimensional vector including a time series relationship is generated through a feature fusion algorithm.

[0112] Step 602: Retrieve and match historical optical compensation cases in the large language model based on the coupling feature vector, and extract a case set from the historical optical compensation cases whose similarity to the current coupling feature vector reaches a preset similarity threshold.

[0113] In step 602, the historical optical compensation cases refer to the set of optimal matching solutions for environment-material-optical parameters learned by the large language model during training. The current coupled feature vector refers to the feature vector generated by real-time binding of the currently collected surface hydrophobic property data (such as the change rate of coating contact angle) and the currently detected environmental dynamic parameters (such as salt fog concentration gradient). Relationship with the coupled feature vector: The coupled feature vector is a general concept, while the current coupled feature vector specifically refers to the instantiated vector generated in real time during this processing. The case set refers to the group of successful optical compensation cases retrieved from the historical database of the large language model that match the current environment-material characteristics. Each case contains the corresponding relationship between specific environmental parameters, material properties, and the optimal optical compensation scheme. This set is composed by calculating the similarity scores between the current coupled feature vector and historical cases, and screening cases with a score exceeding a threshold (such as 0.85) for guiding the generation of the current strategy. For example, for the oil pressure gauge scenario, 3 historical optimal cases with similar salt fog gradients (10-20% / cm) and contact angle change rates (4-6 degrees / hour) may be retrieved.

[0114] In the embodiment of the present application, the coupled feature vector is used as a query condition to perform nearest neighbor search in the case library of the large language model, and successful compensation cases with a similarity exceeding the preset threshold are returned.

[0115] Step 603: According to the case set, calculate the predicted value of the droplet distribution density and the predicted value of the salt crystal deposition thickness on the surface of the anti-corrosion coating of the marine instrument under the current environmental conditions.

[0116] In step 603, the current environmental conditions specifically refer to the actual environmental parameters (such as instantaneous salt fog concentration, humidity, etc.) detected during the execution of this recognition task, which are used to dynamically generate predicted values. The predicted value of the droplet distribution density and the predicted value of the salt crystal deposition thickness refer to the distribution state of the interference substances that may be formed in the current environment based on historical cases.

[0117] In the embodiment of the present application, according to the environmental material relationship in the matching cases, combined with the current parameter differences, linear deduction is performed to calculate the predicted distributions of droplets and salt crystals.

[0118] Step 604: Generate corresponding light source wavelength adjustment suggestion information and polarization angle adjustment suggestion information according to the predicted value of the droplet distribution density.

[0119] In step 604, the light source wavelength adjustment suggestion information and the polarization angle adjustment suggestion information refer to the optical parameter optimization scheme required for predicting the droplet distribution.

[0120] In the embodiment of the present application, according to the droplet density distribution characteristics, a light source wavelength that can highlight the edge contrast of the droplets is selected, and the best polarization angle for suppressing the surface reflection of the droplets is calculated.

[0121] Step 605: Generate corresponding filter transmittance curve adjustment suggestion information according to the predicted salt crystal deposition thickness value.

[0122] In step 605, the filter transmittance curve adjustment suggestion information refers to a filter scheme designed for the salt crystal deposition characteristics.

[0123] In the embodiment of the present application, analyze the influence of the salt crystal thickness on the scattered light, and generate a transmittance curve that can effectively filter the scattered noise.

[0124] Step 606: Synthesize the light source wavelength adjustment suggestion information, the polarization angle adjustment suggestion information, and the filter transmittance curve adjustment suggestion information to generate an optical compensation strategy.

[0125] In the embodiment of the present application, perform conflict detection and parameter fusion on the wavelength, polarization, and filter suggestions to generate a final compensation instruction set.

[0126] The following is a specific example: Taking the identification of the engine tachometer "1500 RPM" as an example, it is measured that the coating contact angle decreases at a rate of 8 degrees per hour from the initial 125 degrees (the change rate is calculated through three consecutive measurement values of 125 degrees, 119 degrees, and 112 degrees), and at the same time, the salt fog concentration gradient increases by 20% per centimeter (calculated from the concentration difference measured by the upper and lower sensors). After binding these two parameters according to the time stamp to generate a coupled feature vector, 2 similar cases (similarity 0.88) are retrieved from the large language model case library: Case 1 uses a wavelength of 455 nm and a polarization angle of 55 degrees, and Case 2 uses a wavelength of 440 nm and a polarization angle of 65 degrees. According to the deduction of the case data, it is predicted that the current droplet distribution density is 15 - 18 per square millimeter (the weighted average of 12 in Case 1 and 20 in Case 2), and the salt crystal deposition thickness is 0.05 mm (interpolated calculation according to the concentration gradient). Comprehensively, it is recommended to select a wavelength of 450 nm (the average of 455 nm and 440 nm) and a polarization angle of 60 degrees (the intermediate value of 55 degrees and 65 degrees), and the filter transmittance curve is set to have a light transmittance of 85% at 450 nm (calculated according to the inverse relationship between the salt crystal thickness and the light transmittance).

[0127] In the embodiment of the present application, through the above steps, the intelligent generation of the optical compensation strategy is realized, effectively solving the problem of difficult adaptation of optical parameters in the rain and fog environment, and improving the character recognition stability under different environmental conditions.

[0128] To solve the problem of poor imaging quality of the instrument in the rain and fog environment and further improve the feature discrimination of the interference fringe pattern, in some embodiments, step 103: Generating the interference fringe pattern corresponding to the surface of the anti-corrosion coating of the marine instrument according to the optical compensation strategy includes: Step 701: Based on the light source wavelength adjustment recommendation information in the optical compensation strategy, adjust the wavelength of the incident beam of the illumination light source so that the incident beam excites a scattered characteristic beam representing the droplet morphology on the surface of the anti-corrosion coating.

[0129] In step 701, the scattered characteristic beam refers to the beam formed after the incident light is scattered at the droplets on the surface of the anti-corrosion coating, and its intensity distribution reflects the size and distribution density of the droplets.

[0130] In the embodiment of the present application, configure a multi-spectral light source according to the wavelength adjustment recommendation so that the wavelength of the output beam adapts to the current droplet curvature, and excite a scattered beam with obvious edge characteristics on the coating surface.

[0131] Step 702: Based on the polarization angle adjustment recommendation information in the optical compensation strategy, adjust the polarization angle of the circularly polarized light source to suppress the stray reflected light caused by the surface salt crystal deposition of the anti-corrosion coating of the marine instrument.

[0132] In step 702, the stray reflected light refers to the interference light formed by the disordered reflection on the surface of the salt crystal particles, and its polarization characteristics are different from those of the target signal.

[0133] In the embodiment of the present application, rotate the transmission axis direction of the circular polarizer according to the polarization angle recommendation to selectively suppress the salt crystal reflected light in a specific direction and retain the useful refraction signal.

[0134] Step 703: Based on the filter transmittance curve adjustment recommendation information in the optical compensation strategy, adjust the filter transmittance curve of the filter array to enhance the interference contrast between the scattered characteristic beam and the refracted characteristic beam.

[0135] In step 703, the refracted characteristic beam is formed by the refraction of the adjusted incident beam passing through the droplets on the surface of the anti-corrosion coating. When the incident beam irradiates the droplets at a specific wavelength and polarization angle, the beam refracts at the droplet-air interface, and its refraction angle and phase change carry the droplet morphology information, forming a refracted characteristic beam with a specific propagation direction. The interference contrast refers to the degree of light and dark difference of the fringes formed by the interference of the scattered light and the refracted light, which affects the feature identifiability.

[0136] In the embodiment of the present application, dynamically configure the transmittance of the multi-band filter according to the filter adjustment recommendation to enhance the target characteristic band while suppressing the noise band.

[0137] Step 704: Through the interaction of the scattered characteristic beam and the refracted characteristic beam on the imaging plane, form an interference light intensity distribution, and generate an interference fringe pattern according to the interference light intensity distribution.

[0138] In step 704, the interference light intensity distribution refers to the spatial intensity change pattern formed by the superposition of two characteristic light beams on the imaging plane. The process of generating the interference fringe pattern: The spatial distribution of the light intensity formed by the interference of the scattered characteristic light beam and the refracted characteristic light beam on the imaging plane is collected by an image sensor, and the interference light intensity values of each pixel are quantized into gray levels to generate a two-dimensional gray distribution pattern that simultaneously encodes the scattering intensity and the refractive phase difference. Specific example: When identifying a marine engine tachometer showing "RPM 1500", the refracted characteristic light beam of the droplet will form concentric interference fringes at the digital edge, while the scattered characteristic light beam of the salt crystal generates radial stripes around the unit symbol "RPM". In the finally generated interference fringe pattern: The vertical stroke of the digit "1" presents parallel stripes with uniform spacing, and the curved part of the letter "P" shows gradually changing curved stripes, and this characteristic distribution directly corresponds to the geometric structure of the tachometer characters.

[0139] In the embodiment of the present application, the light intensity change in the interference region is collected by a high-sensitivity image sensor, and the intensity information distributed in space is quantized into a gray-scale image to generate an interference pattern including droplet and salt crystal characteristics.

[0140] The following is a specific example: Taking the identification of the engine tachometer "1500 RPM" as an example, according to the optical compensation strategy, the light source wavelength is set to 450 nm (calculated from the average curvature radius of the droplet of 4 mm through the formula: optimal wavelength = curvature radius × refractive index difference / 2), the polarization angle is adjusted to 60 degrees (the optimal suppression angle determined through the salt crystal reflection characteristic experiment), and the light transmittance of the filter in the 450 nm band is set to 90% (determined according to the relationship curve between the salt crystal thickness of 0.1 mm and the light transmittance). After implementing the compensation, the incident light beam excites clear scattered characteristic light beams on the instrument surface (parallel stripes with a spacing of 2.1 mm are formed at the digit "1", stripe spacing = wavelength × droplet curvature radius / refractive index difference), and at the same time, the polarization filter effectively suppresses the salt crystal stray light around the letter "R". The refracted characteristic light beam generates circular interference fringes (radius 3.2 mm, calculated from the droplet curvature radius and the incident angle) at the digit "0", and the pattern formed by the interference of the two. The refractive index difference is taken as 1.33 (the refractive index difference between water and air), and the 2.3-fold enhancement coefficient is obtained by comparing the signal-to-noise ratios of the images before and after compensation.

[0141] In the embodiment of the present application, through the above steps, the efficient optical extraction of the characteristics of the rain and fog environment is realized, and the generated interference fringe pattern improves the characteristic distinguishability between droplets and salt crystals, providing high-quality input for subsequent character recognition.

[0142] Figure 2 The structural schematic diagram of a rain and fog weather instrument character recognition system incorporating the prior knowledge of the large model provided by the embodiment of the present application is as Figure 2 shown, and the system includes: An acquisition module 21, configured to obtain surface hydrophobicity characteristic data and environmental dynamic parameters of an anti-corrosion coating of a marine instrument when the marine instrument is in a rain and fog environment.

[0143] An association analysis module 22, configured to perform association analysis on the surface hydrophobicity characteristic data and the environmental dynamic parameters through a pre-trained large language model to generate an optical compensation strategy.

[0144] A first generation module 23, configured to generate an interference fringe pattern corresponding to the surface of the anti-corrosion coating of the marine instrument according to the optical compensation strategy.

[0145] A second generation module 24, configured to perform double-branch processing on the interference fringe pattern to respectively generate a noise suppression pattern and a distortion correction pattern.

[0146] A spatial association module 25, configured to perform spatial association on the noise suppression pattern and the distortion correction pattern to obtain a spatial association pattern, and iteratively optimize the spatial association pattern in combination with the character stroke continuity prior rule in a preset large model prior knowledge base to generate an instrument character recognition result.

[0147] Figure 2 The rain and fog weather instrument character recognition system integrating large model prior knowledge can execute Figure 1 The rain and fog weather instrument character recognition method integrating large model prior knowledge as described in the illustrated embodiment, and its implementation principle and technical effects will not be elaborated here. For the rain and fog weather instrument character recognition system integrating large model prior knowledge in the above embodiment, the specific manners in which each module and unit perform operations have been described in detail in the embodiment related to the method, and will not be elaborated here.

[0148] In a possible design, Figure 2 The rain and fog weather instrument character recognition system integrating large model prior knowledge in the illustrated embodiment can be implemented as a computing device, such as Figure 3 shown, and the computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, and among them, the one or more computer instructions are called and executed by the processing component 32.

[0149] The processing component 32 is the above Figure 1 The rain and fog weather instrument character recognition method integrating large model prior knowledge in the illustrated embodiment.

[0150] Among them, 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 methods. Of course, the processing component may also be implemented by 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 for executing the above methods.

[0151] The storage component 31 is configured to store various types of data to support the operation of the terminal. The storage component can be implemented by any type of volatile or non-volatile storage 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 disks or optical discs.

[0152] Of course, the computing device may also necessarily include other components, such as input / output interfaces, display components, communication components, etc.

[0153] The input / output interface provides an interface between the processing component and the peripheral interface module, and the above peripheral interface module may be an output device, an input device, etc.

[0154] The communication component is configured to facilitate wired or wireless communication between the computing device and other devices, etc.

[0155] Among them, the computing device may be a physical device or an elastic computing host provided by a cloud computing platform, etc. At this time, the computing device may refer to a cloud server, and the above processing component, storage component, etc. may be basic server resources leased or purchased from a cloud computing platform.

[0156] The embodiments of the present application also provide a computer storage medium storing a computer program, and when the computer program is executed by a computer, the above-mentioned Figure 1 rain and fog weather instrument character recognition method integrating large model prior knowledge shown in the embodiments can be implemented.

[0157] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0158] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0159] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0160] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for recognizing instrument characters in rainy and foggy weather by integrating prior knowledge of large models, characterized in that, Including: When the navigation instrument is in a rain and fog environment, obtain the surface hydrophobic property data of the anti-corrosion coating of the navigation instrument and the environmental dynamic parameters; Perform correlation analysis on the surface hydrophobic property data and the environmental dynamic parameters through a pre-trained large language model to generate an optical compensation strategy; Generate an interference fringe pattern corresponding to the surface of the anti-corrosion coating of the navigation instrument according to the optical compensation strategy; Perform double-branch processing on the interference fringe pattern to generate a noise suppression pattern and a distortion correction pattern respectively; Perform spatial correlation on the noise suppression pattern and the distortion correction pattern to obtain a spatial correlation pattern, and iteratively optimize the spatial correlation pattern in combination with the character stroke continuity prior rule in the preset large model prior knowledge base to generate an instrument character recognition result.

2. The method according to claim 1, wherein The iteratively optimizing the spatial correlation pattern in combination with the character stroke continuity prior rule in the preset large model prior knowledge base to generate an instrument character recognition result includes: Extract candidate line segments related to the stroke direction of the instrument characters from the spatial correlation pattern; Match the candidate line segments with the character stroke continuity prior rule to generate a set of candidate line segments; Iteratively optimize the spatial correlation pattern according to the set of candidate line segments to generate an optimized spatial correlation pattern; Extract candidate character boundaries that match the predefined character structure template in the character stroke continuity prior rule from the optimized spatial correlation pattern to generate an instrument character recognition result.

3. The method according to claim 2, wherein The matching the candidate line segments with the character stroke continuity prior rule to generate a set of candidate line segments includes: Divide the line segment direction type set according to the extension direction and length ratio of the candidate line segments in the spatial correlation pattern; Extract the stroke turning angle range corresponding to the line segment direction type set from the character stroke continuity prior rule; Group the candidate line segments into candidate line segment clusters according to the corresponding extension direction; Compare the head and tail endpoint spacing of adjacent line segment pairs in the candidate line segment cluster with a preset spacing tolerance, where the head and tail endpoint spacing is the spacing between the head endpoint of one candidate line segment and the tail endpoint of another candidate line segment in the adjacent line segment pair; For each adjacent line segment pair with a head and tail endpoint spacing less than the spacing tolerance, calculate the angle fluctuation amplitude of the adjacent line segment pair based on the stroke turning angle range; Generate a set of candidate line segments based on the angle fluctuation amplitude.

4. The method according to claim 3, characterized in that, The generating a set of candidate line segments based on the angle fluctuation amplitude includes: Extract the allowable curvature change threshold corresponding to the line segment direction type set from the character stroke continuity prior rule; Eliminate adjacent line segment pairs with an angle fluctuation amplitude exceeding the allowable curvature change threshold within the candidate line segment cluster to generate a target line segment cluster; Taking the center point of the target line segment cluster as a reference, extend a virtual connection line along the curvature change direction, and judge whether the virtual connection line is consistent with the stroke extension trend of the character structure in the character stroke continuity prior rule; When the judgment is consistent, dynamically adjust the angle threshold of adjacent line segments within the target line segment cluster. Connect and integrate adjacent line segment pairs that meet the dynamically adjusted included angle threshold to generate a candidate line segment set.

5. The method according to claim 2, wherein Iteratively optimizing the spatial association map according to the candidate line segment set to generate an optimized spatial association map includes: Dividing a pixel confidence enhancement region associated with the overlapping probability of character strokes in the spatial association map according to the spatial distribution relationship between the endpoints of the line segments in the candidate line segment set; Based on the stroke connection topological constraint relationship in the prior rule of character stroke continuity, perform two-way weighted diffusion on the pixel confidence enhancement region to construct a confidence distribution network covering the complete character stroke contour; According to the confidence distribution network and the candidate line segment set, dynamically enhance or attenuate the confidence level of the pixel points in the spatial association map to generate an optimized spatial association map.

6. The method according to claim 1, characterized in that, Performing correlation analysis on the surface hydrophobic property data and the environmental dynamic parameters through a pre-trained large language model to generate an optical compensation strategy, including: Binding the coating contact angle change rate in the surface hydrophobic property data and the salt fog concentration gradient in the environmental dynamic parameters in a time series to generate a coupled feature vector of the environment and the material; Retrieving and matching historical optical compensation cases in the large language model based on the coupled feature vector, and extracting a case set from the historical optical compensation cases whose similarity to the current coupled feature vector reaches a preset similarity threshold; According to the case set, calculating the predicted value of the droplet distribution density and the predicted value of the salt crystal deposition thickness on the surface of the anti-corrosion coating of the navigation instrument under the current environmental conditions; Generating corresponding light source wavelength adjustment suggestion information and polarization angle adjustment suggestion information according to the predicted value of the droplet distribution density; Generating corresponding filter transmittance curve adjustment suggestion information according to the predicted value of the salt crystal deposition thickness; Integrating the light source wavelength adjustment suggestion information, the polarization angle adjustment suggestion information, and the filter transmittance curve adjustment suggestion information to generate an optical compensation strategy.

7. The method according to claim 1, characterized in that Generating an interference fringe map corresponding to the surface of the anti-corrosion coating of the navigation instrument according to the optical compensation strategy, including: Adjusting the wavelength of the incident beam of the illumination light source based on the light source wavelength adjustment suggestion information in the optical compensation strategy, so that the incident beam excites a scattered characteristic beam representing the droplet morphology on the surface of the anti-corrosion coating; Adjusting the polarization angle of the circularly polarized light source based on the polarization angle adjustment suggestion information in the optical compensation strategy to suppress the stray reflected light caused by the salt crystal deposition on the surface of the anti-corrosion coating of the navigation instrument; Adjusting the filter transmittance curve of the filter array based on the filter transmittance curve adjustment suggestion information in the optical compensation strategy to enhance the interference contrast between the scattered characteristic beam and the refracted characteristic beam; Form an interference light intensity distribution through the interaction of the scattered characteristic beam and the refracted characteristic beam on the imaging plane, and generate an interference fringe map according to the interference light intensity distribution.

8. A rain and fog weather instrument character recognition system integrating prior knowledge of large models, characterized in that, Including: An acquisition module for acquiring surface hydrophobic property data and environmental dynamic parameters of the anti-corrosion coating of the navigation instrument when the navigation instrument is in a rain and fog environment; The correlation analysis module is used to perform correlation analysis on the surface hydrophobic property data and the environmental dynamic parameters through a pre-trained large language model to generate an optical compensation strategy; The first generation module is used to generate an interference fringe pattern corresponding to the surface of the anti-corrosion coating of the marine instrument according to the optical compensation strategy; The second generation module is used to perform double-branch processing on the interference fringe pattern to respectively generate a noise suppression pattern and a distortion correction pattern; The spatial correlation module is used to perform spatial correlation on the noise suppression pattern and the distortion correction pattern to obtain a spatial correlation pattern, and iteratively optimize the spatial correlation pattern in combination with the character stroke continuity prior rule in the preset large model prior knowledge base to generate an instrument character recognition result.

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 a method for recognizing instrument characters in rainy and foggy weather conditions that integrates prior knowledge of a large model 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, it implements a method for recognizing instrument characters in rainy and foggy weather conditions that integrates prior knowledge of a large model as described in any one of claims 1 to 7.

Images