Noise-resistant temperature measurement laser micro-dot code decoding method and system for curved copper and aluminum fittings

Abstract

This invention relates to the field of digital identification technology for power equipment, and discloses a noise-resistant temperature-measuring laser micro-dot code decoding method and system for curved copper and aluminum fittings. The method includes: constructing an observation coordinate system and performing polarization analysis to obtain a set of polarization features; using a wave-particle duality adjustment factor to perform energy decoupling to obtain a wave-like energy distribution field; applying a phase angle distribution set to perform directional spatial filtering to generate a texture feature map and extracting quality assessment parameters; using an AI image-locking model to lock the geometric feature set based on the quality assessment parameters, and using an AI polymer image-locking model to refine the segmentation to obtain a set of candidate code points; reconstructing a micro-dot logic array based on the local radius of curvature; constructing a topological error-correction map and performing feature compensation to obtain a full-dimensional array, and outputting a logic verification pass identifier. This invention solves the problems of high reflectivity semantic aliasing and manifold distortion, and achieves reliable reading and closed-loop identity verification of micro-dot codes in harsh industrial environments.

Description

Technical Field

[0001] This invention relates to the field of digital identification technology for power equipment, and more specifically, to a noise-resistant temperature-measuring laser micro-dot code decoding method and system for curved copper and aluminum fittings. Background Technology

[0002] With the evolution of the power Internet of Things, real-time monitoring of the operating status of high-voltage cables has become crucial for ensuring power grid safety. Currently, copper-aluminum fittings with built-in temperature-sensing ceramic chips, known as temperature-sensing wells, are commonly used for contact temperature measurement, closely attached to the cable. To achieve accurate digital binding between the temperature-sensing node and the cable phase, reliable coding and recognition of the direct component markings on the surface of the temperature-sensing well are necessary. Because these cable temperature-sensing fittings often have small-radius cylindrical surfaces or complex irregular curved surfaces, and the materials themselves possess high reflectivity and are prone to oxidation, coupled with the harsh industrial environment such as cable trenches, reliable recognition of laser micro-dot codes has become a bottleneck restricting industry automation. However, existing DPM decoding algorithms are mainly designed based on the quasi-planar assumption and grayscale contrast features. When processing curved copper-aluminum fittings, they generally face bottlenecks such as "drastic fluctuations in imaging signal-to-noise ratio" and "feature dimension collapse," making it difficult to achieve both decoding success rate and robustness under complex operating conditions. Existing technologies mostly rely on macroscopic processing methods such as global or local dynamic threshold segmentation, neglecting the optical-semantic coupling effect between the specular reflection mechanism of copper-aluminum materials and laser-etched textures. Under strong industrial light or multi-source interference, specular reflection noise from metal surfaces and micro-etched dot lattices can easily induce semantic aliasing in image space. This means that saturated noise pixels in bright reflective areas exhibit a high degree of similarity in grayscale distribution to valid code points, making it impossible for traditional algorithms to uniquely determine the existence of code points through single-modal grayscale projection, leading to frequent misjudgments or omissions. Furthermore, oxide layer wear and oil residue on hardware surfaces introduce structural noise at the same scale as micro-dot codes. Existing technologies primarily focus on static single-pixel noise reduction, lacking in-depth modeling of code point topological relationships and physical reflection models, making it difficult to identify weak contrast features implicitly obscured by noise. For curved surfaces, traditional algorithms typically use linear mapping models such as affine or perspective transformations for geometric correction, which struggles to characterize the manifold distortion in feature space caused by nonlinear curvature. In scenarios where the micro-dot code assignment area is limited and the carrier radius is extremely small, code points in edge regions experience severe perspective collapse and shape shrinkage due to the multiple couplings between surface curvature and the acquisition perspective, resulting in semantic shifts in the spatial topological relationships between code elements. Current technologies mostly focus on surface correction of the overall image, neglecting the interference of local curvature fluctuations on the geometric centroid positioning of code points. This leads to the failure of error correction mechanisms due to the lack of dimensionality in low-level feature acquisition, causing the system to encounter decoding bottlenecks caused by geometric mapping distortion. Therefore, how to shift from static grayscale thresholding to dynamic physical property enhancement, transforming macroscopic planar geometric correction into precise orientational reconstruction of surface topological trajectories to eliminate semantic aliasing under high reflectivity and compensate for nonlinear distortion of the surface, is a technical challenge to be solved in this field.

[0003] In the prior art, Chinese patent CN102426643A discloses a wheel-type laser coding device and a coding and decoding method. This system includes a curved barcode unit, a rotation drive module, a laser emission and reception component, and a coding processing element, enabling the positioning and decoding of barcodes on curved or rotating surfaces. It uses a motor to drive the curved surface to rotate, converting the reflected light from the barcode into a voltage sequence, which is then analyzed by the coding processing element. This solves the problem that traditional barcodes cannot be applied to curved surfaces, providing a feasible path for laser coding on non-planar carriers. Chinese patent CN121253480A discloses a modulation and demodulation method and system based on continuous data discrete reconstruction. This system adopts a three-layer architecture of "data acquisition, discrete reconstruction, and harmonic extraction," including a laser drive unit, a signal reconstruction module, and a lock-in amplifier decoding unit, which respectively realize drive current generation, light intensity signal optimization, and effective signal extraction. It generates a reference timing sequence through the characteristics of the laser output signal itself, achieving self-matching of the modulation frequency, effectively suppressing external signal interference, and improving the stability of harmonic extraction.

[0004] However, while the two existing technologies mentioned above have some value in the interpretation of barcodes on curved surfaces and in anti-interference signal processing, they fail to address the core pain points in the decoding of laser micro-dot codes on curved copper and aluminum fittings: "static surface adaptation, metal reflectivity noise reduction, and accurate micro-dot code recognition." Specifically, the Chinese patent with publication number CN102426643A focuses on barcode interpretation on rotating surfaces, relying on a motor to drive the surface rotation to complete signal acquisition, which is unsuitable for static curved surfaces like copper and aluminum fittings. Furthermore, its barcodes are sequences of color blocks with varying reflectivity, not micro-dot code structures, and it lacks an anti-noise mechanism designed for the strong reflectivity of metal surfaces, making it difficult to suppress signal distortion caused by scattering noise. The Chinese patent with publication number CN121253480A focuses on anti-interference demodulation for mining laser gas detection. Its noise reduction design is tailored to signal interference scenarios in gas detection, but it does not consider the distortion problem of laser micro-dot codes on metal surfaces and lacks a targeted algorithm for extracting micro-dot code features. Neither of them has established a noise-resistant decoding system for metal static curved surface micro-dot codes. They cannot cope with the code image distortion caused by the large curvature surface of copper and aluminum fittings, nor can they eliminate the decoding error caused by metal reflection. They cannot meet the high-precision traceability requirements of laser micro-dot codes in scenarios such as cable temperature measuring wells equipped with temperature measuring ceramic chips. Summary of the Invention

[0005] This invention is applicable to high-voltage cable operation status monitoring and intelligent inspection scenarios, and can meet the digital identification and reading requirements of curved copper and aluminum fittings with high reflectivity and complex geometric features in harsh industrial environments. Through a wave-particle duality energy decoupling model, the specular reflection noise dominated by particle properties and the diffuse reflection signal dominated by wave properties are physically separated, eliminating the occlusion of texture features in the laser micro-dot code region by semantic aliasing from the bottom layer. The AI ​​image-locking model dynamically adjusts the judgment sensitivity using quality assessment parameters, achieving adaptive and accurate locking of the geometric feature set of the region to be decoded under oxidative wear background. The AI ​​polymer image-locking model uses gravitational weights and geometric contraction stress to perform pixel-level feature stripping in the dynamic mapping space, effectively solving the problem of symbol edge adhesion caused by small radius curvature. A reconstructed weight distribution set, in conjunction with a digital standard grid, performs spatial transformation, restoring the topologically misaligned candidate code point set to a micro-dot logic array. The topology error correction map performs feature dimension compensation based on correlation probability, and combined with cyclic redundancy check, ensures that the output result is highly faithful to the physical entity information. This invention significantly improves the robustness and success rate of micro-dot code decoding under complex working conditions, and realizes full life-cycle identity traceability and reliable supervision of cable temperature measurement nodes.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A noise-resistant laser micro-dot code decoding method for curved copper and aluminum fittings includes: The original image set of the curved copper-aluminum hardware surface is obtained. Polarization analysis is performed on the original image set to obtain a polarization feature set. Energy decoupling is performed on the polarization feature set to obtain a wave energy distribution field. Directional spatial filtering is performed on the wave energy distribution field to generate a texture feature map. Feature calculation is performed on the texture feature map to obtain quality assessment parameters. Based on the quality assessment parameters, an AI-based image capture model is constructed and applied to obtain the target recognition confidence value. The target recognition confidence value is then binarized to obtain a geometric feature set. Based on the geometric feature set, an AI-based polymer image capture model is constructed and applied to obtain a candidate code point set. The candidate code point set is then spatially mapped to obtain a micro-dot logic array. The logic nodes to be detected are determined based on the micro-point logic array. Spatial topology extraction is performed on the logic nodes to be detected to obtain the topology error correction map. Feature compensation is performed on the topology error correction map to obtain the logic verification pass mark.

[0007] Furthermore, the method for obtaining the polarization feature set includes: The imaging sensor is used to collect information on the intensity of reflected light with different polarization angles, and a two-dimensional brightness distribution matrix consisting of multiple pixel sampling points is generated, which is the original image set. An observation coordinate system is constructed based on the original image set. For each pixel sampling point in the original image set, a parameter distribution set is parsed out. The parameter distribution set includes a total light intensity parameter representing macroscopic brightness, a horizontal linear polarization parameter representing the difference between horizontal and vertical vibration, and a 45-degree linear polarization parameter representing the difference between oblique vibration. Physical parameter mapping is performed based on the parameter distribution set to generate a polarization degree distribution set that characterizes the proportion of linear polarization energy, and a phase angle distribution set that characterizes the rotation angle of the reflected light vector vibration direction relative to the observation coordinate system. The polarization degree distribution set and the phase angle distribution set are then combined and encapsulated into a polarization feature set.

[0008] Furthermore, the wave-like energy distribution field includes: A wave-particle duality adjustment factor is introduced, and the energy weight corresponding to each pixel sampling point is calculated by using the numerical values ​​of the polarization degree distribution set and the wave-particle duality adjustment factor. The total light intensity parameter in the parameter distribution set is multiplied by the energy weight at the corresponding position using pixel-level arithmetic multiplication. The energy intensity of the wave component dominated by diffuse reflection in the reflected light field is extracted to obtain the wave energy characteristic value. The wave energy characteristic values ​​corresponding to all pixel sampling points on the photosensitive chip are mapped to a two-dimensional digital set, which is defined as the wave energy distribution field.

[0009] Furthermore, the method for obtaining the quality assessment parameters includes: The phase gradient values ​​within a preset local spatial window are calculated using the phase angle distribution set, and a convolution kernel weight matrix is ​​generated. An adaptive convolution operation is performed on the wave energy distribution field using the convolution kernel weight matrix to obtain the filtered response value; Pixels whose filtered response values ​​are greater than or equal to a preset response threshold are extracted as contour mapping points, and texture feature maps are generated by connecting the contour mapping points. Obtain the spatial coordinates of the contour mapping points in the observation coordinate system, extract the values ​​of the corresponding spatial coordinate positions from the wave energy distribution field to form a code point feature set, and extract the values ​​of pixel positions other than the corresponding spatial coordinate positions to form a background feature set. Calculate the mean of the code point feature set and the standard deviation of the background feature set, and then perform an arithmetic division operation between the two to obtain the quality assessment parameters.

[0010] Furthermore, the method for obtaining the geometric feature set includes: Set a recognition threshold weight, perform an arithmetic product operation between the quality assessment parameter and the recognition threshold weight, and obtain the interest region recognition threshold. Build and apply an AI-based image capture model to perform a sliding search on the texture feature map and output the target recognition confidence score. When the target recognition confidence score is greater than or equal to the region of interest recognition threshold, the region of interest is locked. Obtain the center coordinates of the region of interest in the observation coordinate system, and use the phase angle distribution set to fit and calculate the local radius of curvature corresponding to the region of interest; The region of interest, center coordinates, and local radius of curvature are digitally encapsulated to obtain a geometric feature set.

[0011] Furthermore, the method for obtaining the candidate codepoint set includes: Based on the geometric feature set, a grid array corresponding to the pixel distribution of the region of interest is constructed as a dynamic mapping space. Each node in the grid array is defined as a single unit, and each single unit is filled with texture feature values. The balance distance between individual units is determined by using a preset etching spacing, and the gradient components of the texture feature values ​​are calculated. Calculate the geometric contraction stress and gravitational weight of a single unit; whereby the geometric contraction stress is determined based on the degree to which the real-time displacement distance of adjacent single units deviates from the equilibrium distance, and the gravitational weight is determined based on the gradient component and the local radius of curvature. Based on the vector synthesis result of geometric contraction stress and gravitational weight, the single unit is driven to shift its position in the observation coordinate system, generating a dynamic arrangement state; By recursively retrieving and labeling all individual units within the dynamic arrangement state, multiple mutually independent physically connected domains are generated, which are defined as candidate code points. All candidate code points are then aggregated to obtain a set of candidate code points, which includes the barycenter coordinates of the code points in the observation coordinate system and the code point feature values.

[0012] Furthermore, the method for obtaining the micro-dot logic array includes: Set the focal length compensation coefficient, and calculate the reconstruction weight distribution set for each candidate code point in the candidate code point set based on the local radius of curvature, center coordinates and focal length compensation coefficient. The reconstruction weight distribution set includes horizontal reconstruction weight and vertical reconstruction weight. The code centroid coordinates are weighted using horizontal and vertical reconstruction weights to generate reconstruction coordinate pairs. Obtain the coding scale parameter that characterizes the original physical structure properties of the laser micro-dot code region, and construct a digital standard grid in the observation coordinate system based on the coding scale parameter and the etching spacing. The digital standard grid contains multiple logical nodes. Calculate the Euclidean distance between the reconstructed coordinate pair and the logical node, and assign the reconstructed coordinate pair to the logical node with the smallest Euclidean distance; The binary logic level is determined based on the allocation status of the logic nodes, and the binary logic levels of all logic nodes are combined in the row and column order of the rectangular array to obtain the micro-dot logic array.

[0013] Furthermore, the topological error correction map includes: Extract nodes in the micro-dot logic array whose binary logic level is zero, and define them as logic nodes to be detected; Search the eight-neighbor range of the logic node to be detected within the digital standard grid, and extract the nodes with a binary logic level of one to form an auxiliary reference point set; The distance distribution probability of the logic node to be detected relative to the auxiliary reference point set is calculated based on the etching spacing, and the direction distribution probability of the logic node to be detected is calculated based on the orthogonal arrangement characteristics of the digital standard grid. The distance distribution probability and the direction distribution probability are linearly weighted and superimposed to obtain the logical recovery probability value. The logical recovery probability values ​​corresponding to all logical nodes to be detected constitute the topology error correction map.

[0014] Furthermore, the logical verification is achieved through the following identifiers: To assist in the distribution of node confidence weight coefficients in the reference point set, the confidence weight coefficients are used to perform a weighted operation on the logical recovery probability values ​​in the topology error correction graph to obtain the feature dimension compensation factor of the logical node to be detected. The feature dimension compensation factor is compared with the preset recombination threshold. If the feature dimension compensation factor is greater than or equal to the recombination threshold, the binary logic level of the logic node to be detected is corrected to one to obtain a full-dimensional array. Set an extraction scan path, construct the bit stream to be verified from the full-dimensional array according to the extraction scan path, and perform cyclic redundancy check operation on the bit stream to be verified. If the remainder of the cyclic redundancy check operation is zero, a logical check pass flag is obtained.

[0015] A noise-resistant temperature-sensing laser micro-dot code decoding system for curved copper and aluminum fittings, used to implement the aforementioned noise-resistant temperature-sensing laser micro-dot code decoding method for curved copper and aluminum fittings, the system comprising: Energy decoupling module: used to acquire the original image set of the curved copper and aluminum hardware surface, perform polarization analysis operation on the original image set to obtain the polarization feature set, perform energy decoupling on the polarization feature set to obtain the wave energy distribution field, perform directional spatial filtering on the wave energy distribution field to generate a texture feature map, and perform feature calculation on the texture feature map to obtain quality assessment parameters; Manifold Reconstruction Module: This module is used to construct and apply an AI-based image-locking and grasping model based on quality assessment parameters, obtain target recognition confidence values, perform binarization judgment on the target recognition confidence values ​​to obtain a geometric feature set, construct and apply an AI polymer image-locking model based on the geometric feature set to obtain a candidate code point set, and perform spatial mapping on the candidate code point set to obtain a micro-point logic array. Topology correction module: used to determine the logic node to be detected based on the micro-point logic array, perform spatial topology extraction on the logic node to be detected to obtain the topology correction map, perform feature compensation on the topology correction map to obtain the logic verification pass mark.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves effective separation of specular reflection saturation noise dominated by particle properties and diffuse reflection light signals dominated by wave properties at the physical energy level by constructing a wave-particle duality energy decoupling model and a polarization feature set. This solves the problem of features being submerged by noise due to semantic aliasing in traditional imaging algorithms. The AI ​​polymer image-locking model utilizes the dynamic balance logic of gravitational weights and geometric contraction stress to achieve pixel-level precise peeling of micro-etched pits that are adhered by small radius curvature compression. This solves the technical problem of code centroid extraction deviation in traditional DPM decoding algorithms under complex curved surfaces. The reconstructed weight distribution set collaborative topology error correction map restores the distorted candidate code point set to a micro-point logic array. It also uses feature dimension compensation factors to achieve intelligent retrieval of logic levels in surface damage areas, significantly improving the robustness and success rate of decoding in harsh power field environments. This ensures the physical consistency of identification information reading of cable temperature measuring wells equipped with temperature measuring ceramic chips and solves the technical problem of difficulty in reading temperature measuring nodes due to hardware reflection and wear. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a noise-resistant temperature measurement laser micro-dot code decoding method for curved copper and aluminum fittings provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of multimodal polarization image acquisition and coordinate system construction provided by an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the relationship between the dynamic mapping space and the topological constraints of a single unit cell, provided in an embodiment of the present invention. Figure 4 This is a functional template diagram of an anti-noise temperature measurement laser micro-dot code decoding system for curved copper and aluminum fittings provided in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1 Please see Figure 1 As shown, this embodiment provides a noise-resistant temperature-sensing laser micro-dot code decoding method for curved copper and aluminum fittings, including: Step S10: Obtain the original image set of the curved copper-aluminum hardware surface, perform polarization analysis operation on the original image set to obtain the polarization feature set, perform energy decoupling on the polarization feature set to obtain the wave energy distribution field, perform directional spatial filtering on the wave energy distribution field to generate a texture feature map, and perform feature calculation on the texture feature map to obtain the quality assessment parameters.

[0021] Further, step S10 includes: Step S11: Obtain the original image set of the curved copper-aluminum hardware surface and construct the observation coordinate system. Perform polarization analysis operation on the original image set based on the observation coordinate system to obtain the polarization feature set.

[0022] In the high-precision traceability and intelligent inspection process of power equipment, temperature measuring wells used to monitor cable temperature are typically made of curved copper-aluminum hardware, which encapsulates a temperature-sensing ceramic chip. The curved copper-aluminum hardware serves as the physical carrier of laser micro-dot codes, possessing a highly reflective metallic surface. Since laser micro-dot codes are a dot matrix structure composed of microscale laser-etched pits, when a light source shines on the surface of the curved copper-aluminum hardware, the specular reflection of light from the surface creates extremely bright blind spots. The saturation noise generated by these blind spots obscures the texture information of the laser micro-dot codes, preventing conventional image acquisition equipment from obtaining clear laser micro-dot code features. To eliminate the interference of specular reflection light on laser micro-dot code recognition, a reference frame capable of signal separation using the polarization characteristics of light waves is established. The aim is to transform passive, interference-prone single-mode brightness monitoring into active, physically distinguishable polarization-dimensional monitoring.

[0023] Specifically, to accurately describe the optical reflection characteristics of the cable temperature sensing well, i.e., the surface of the curved copper-aluminum fitting, a fixed observation coordinate system is established with the physical geometric center of the arc-shaped temperature sensing surface on the curved copper-aluminum fitting used to encapsulate the temperature-sensing ceramic chip as the origin O, based on the imaging projection point. The imaging projection point refers to the geometric mapping point formed on the surface of the image sensor's photosensitive chip after the principal ray emitted from the physical geometric center of the curved copper-aluminum fitting passes through the center of the optical lens of the imaging sensor. Its function is to establish the collinear mapping relationship between the physical spatial position of the curved copper-aluminum fitting and the pixel grid of the imaging sensor. In the observation coordinate system, the imaging optical axis direction of the imaging sensor is defined as the Z-axis; the straight line originating from the origin O and parallel to the pixel row direction is defined as the X-axis; and the straight line originating from the origin O and parallel to the pixel column direction is defined as the Y-axis. See also... Figure 2 This is a schematic diagram of multimodal polarization image acquisition and coordinate system construction provided by an embodiment of the present invention. Figure 2 In the data acquisition scenario shown, the curved copper-aluminum fitting is an industrial metal connector with a cylindrical arc surface. Internally, it encapsulates a temperature-sensing ceramic chip for sensing cable temperature. Multiple laser micro-dot code areas are distributed on the surface of the curved copper-aluminum fitting. Figure 2 An example of a laser micro-dot code region is provided. This region refers to a physical information-carrying field formed on the substrate surface of a curved copper-aluminum fitting using a controlled etching process with a high-energy laser beam. It consists of multiple microscopic etched pits arranged in a predetermined spatial topology. The information carried within this physical information-carrying field uniquely identifies the physical identity of the temperature-sensing ceramic chip. The imaging sensor comprises a photosensitive chip and a housing. The imaging sensor faces the curved copper-aluminum fitting along the Z-axis. An optical lens and a dynamic polarization rotation unit are mounted at the front end of the imaging sensor. The dynamic polarization rotation unit is an optical component that can change the transmission axis direction of the analyzer by adjusting an electronically controlled motor. Its function is to acquire the intensity components of reflected light in different vibration directions. The dynamic polarization rotation unit acquires polarization information at different angles by rotating different polarization observation vectors. Figure 2 The study also demonstrates the reflection separation logic of specular and diffuse reflection paths on the surface of curved copper-aluminum fittings. The origin O, serving as the imaging projection point, is precisely positioned at the geometric center of the image sensor's photosensitive chip surface. The X and Y axes are strictly aligned with the pixel row and column directions, respectively, while the Z axis coincides with the imaging optical axis of the image sensor. Dashed lines illustrate the mapping process by which the physical geometric center of the curved copper-aluminum fitting, after passing through the optical lens, forms the origin O on the photosensitive chip surface.

[0024] After establishing the observation coordinate system, an acquisition array composed of an imaging sensor and a dynamic polarization rotation unit is used to acquire original images of the curved copper-aluminum hardware surface according to a preset polarization observation vector period. The polarization observation vector refers to the vibration direction vector of the analyzer's transmission axis at the moment the imaging sensor acquires the image. The setting is based on extracting the intensity information of reflected light in multiple independent vibration dimensions by changing the physical angle of the analyzer's transmission axis. The number of polarization observation vectors is set to satisfy the solvability constraint of the Stokes parameter matrix linear equation system, that is, by solving the total light intensity component, horizontal polarization component, and oblique polarization component through at least three independent intensity equations, thereby achieving mathematical reconstruction of the polarization state of the reflected light field. For example, it is set to 3; the polarization angles of the polarization observation vectors are uniformly distributed according to a preset angular interval. The angular interval refers to the angular displacement difference between the polarization angles of two adjacent polarization observation vectors. The setup is based on Malus's law and the periodicity of polarization intensity variation with the analyzer rotation angle. By performing equidistant sampling within a 180-degree polarization period, the polarization modulation characteristics of the reflected light field are maximized, and Stokes parameter calculation errors caused by uneven sampling point distribution are eliminated. For example, if the number of polarization observation vectors is 3, the polarization angles are set to 0 degrees, 60 degrees, and 120 degrees respectively. The acquisition array obtains the corresponding number of original images for each polarization observation vector. The original image refers to the two-dimensional brightness distribution matrix generated after the imaging sensor performs photoelectric conversion on the light intensity information reflected from the curved copper-aluminum hardware surface under a specific polarization observation vector. To balance the real-time computational requirements of the algorithm with the analytical completeness of the physical model, the following detailed explanation uses a setting of 3 polarization observation vectors. That is, the acquisition array obtains 3 original images corresponding to the polarization observation vectors, defined as the original image set; the original image set is defined as the first original image I1, the second original image I2, and the third original image I3 using the observation coordinate system, with polarization angles of 0 degrees, 60 degrees, and 120 degrees respectively. The original image set is then analyzed to obtain a parameter distribution set. This parameter distribution set consists of a total light intensity parameter S0, a horizontal linear polarization parameter S1, and a 45-degree linear polarization parameter S2. The construction logic utilizes intensity sampling components under multiple polarization observation vectors to reconstruct physical parameters describing the polarization state of the light field at each pixel coordinate through linear combination, thereby transforming the discrete original image brightness into a feature space with physical meaning. The total light intensity parameter represents the digital value of the macroscopic brightness of the reflected light field on the curved copper-aluminum hardware surface; the formula corresponding to the total light intensity parameter is: The horizontal linear polarization parameter represents the digital value of the difference between the intensity of the vibrational component of the reflected light field in the direction parallel to the X-axis and the intensity of the vibrational component in the direction parallel to the Y-axis. The corresponding formula is: The 45-degree linear polarization parameter is a digital value representing the difference between the intensity of the vibrational component of the reflected light field in the direction of rotation of 45 degrees relative to the X-axis and the intensity of the vibrational component in the direction of rotation of 135 degrees relative to the X-axis. The corresponding formula is: .

[0025] A pixel-level polarization degree distribution set P and a phase angle distribution set A are generated using a parameter distribution set. The polarization degree distribution set is a digitized set composed of the polarization degree values ​​corresponding to each pixel sampling point on the photosensitive chip. The polarization degree values ​​characterize the proportion of linearly polarized energy to the total energy in the reflected light field. The formula for calculating the polarization degree distribution set is: The logic behind constructing the polarization degree distribution set is as follows: Feature differentiation is achieved by utilizing the high polarization maintenance of specularly reflected light and the depolarization characteristics of diffusely reflected light generated by micro-etched pits within the laser micro-dot code region. Specifically, the polarization degree value of specularly reflected light on a smooth substrate surface is in a high score range, for example, 0.8 to 1; the polarization degree value of diffusely reflected light within the laser micro-dot code region is in a low score range, for example, 0.1 to 0.3. The polarization degree distribution set provides a quantitative discrimination basis with spatial location index for removing saturation noise generated by blind spots. The phase angle distribution set refers to the digital set composed of the phase angle values ​​corresponding to each pixel sampling point on the photosensitive chip. The phase angle value represents the rotation angle of the vibration direction of the reflected light vector relative to the X-axis of the observation coordinate system. The calculation formula for the phase angle distribution set is: ,in, The arctangent function is represented here. The logic for constructing the phase angle distribution set is as follows: the vector direction characteristics of the reflected light field at the microscopic scale are restored through the parameter distribution set, with the aim of providing the microscopic geometric normal orientation value of the curved copper-aluminum hardware surface at each pixel point. The polarization degree distribution set and the phase angle distribution set are combined to obtain the polarization feature set.

[0026] Step S12: Use the polarization feature set to perform energy decoupling to obtain the wave energy distribution field.

[0027] After obtaining the polarization feature set, in order to capture the negative superposition effect of energy generated by specular reflection light on the surface of curved copper-aluminum hardware, i.e., the distortion of texture features caused by energy interference between the saturation noise generated by specular reflection light and the diffuse reflection light generated by the laser micro-dot code region on the imaging plane, a logical analytical space for performing physical energy separation is defined in the observation coordinate system using the polarization feature set. This solves the problem of the loss of hidden texture energy caused by the high reflectivity of the metal substrate. In the process of processing the optical information of the surface of curved copper-aluminum hardware, specular reflection light acts as the physical carrier that generates saturation noise and blocks the laser micro-dot code region. The energy distribution characteristics of specular reflection light in the high-speed acquisition transient are strictly constrained by the physical boundary determined by the polarization degree distribution set. Since the polarization feature set provides a high-fidelity physical discrimination benchmark, an energy separation process is established that can map the topological features of the parameter distribution set with the wave-particle duality energy decoupling model. The aim is to transform the single-dimensional grayscale brightness processing in traditional technology into an active, full-pixel field-covering, dynamic energy field analysis with physical energy attribute orientation.

[0028] Specifically, to accurately describe the energy distribution pattern on the surface of curved copper-aluminum fittings, a wave-particle duality adjustment factor B is introduced within the observation coordinate system, using the total light intensity parameter as the original energy benchmark of the reflected light field. This wave-particle duality adjustment factor is an empirical coefficient determined based on the material refractive index of the curved copper-aluminum fitting and the ambient light intensity. The determination method is as follows: Polarization degree distribution sets of copper-aluminum material samples are collected under standard light source illumination. Statistical analysis is performed on the polarization degree values ​​of the specular reflection region and the laser-etched region, respectively. The difference between the mean polarization degree values ​​of the two regions is calculated as the polarization response slope. The product of the polarization response slope and the material refractive index is used as the benchmark value of the wave-particle duality adjustment factor, and linear correction is performed based on the ratio of ambient light intensity to standard light intensity. For example, the value range of the wave-particle duality adjustment factor is set to 0.85 to 1.25. Its function is to adjust the suppression intensity of saturation noise generated by the blind spot in the wave-particle duality energy decoupling model. The energy weight W is calculated using a wave-particle duality adjustment factor and a polarization degree distribution set. This energy weight is a digital distribution set corresponding to the pixel coordinates of the photosensitive chip, representing the proportion of energy of the wave component in the reflected light field. Its function is to provide a physical attenuation factor based on polarization characteristics for the total light intensity parameter, thereby achieving nonlinear suppression of particle-dominated saturation noise. The formula is: The formula is constructed by utilizing the difference in the numerical distribution of the particle properties of specularly reflected light and the wave properties of diffusely reflected light within the polarization degree distribution set to achieve energy mapping. Since the polarization degree value corresponding to specularly reflected light approaches 1, the calculated energy weight approaches 0, thus suppressing the energy contribution of the particle component at the physical level. Conversely, since the polarization degree value corresponding to diffusely reflected light generated by the laser micro-dot code region is low and its distribution is stable, the calculated energy weight W remains in the high-score range, thus preserving the characteristics of the wave component. The wave energy characteristic value is obtained by performing an arithmetic product operation on the total light intensity parameter S0 and the energy weight W at each sampling pixel location within the observation coordinate system, representing the energy intensity of the wave component in the reflected light vector. The digital set of wave energy characteristic values ​​corresponding to each pixel sampling point on the photosensitive chip is defined as the wave energy distribution field. The function of the wave energy distribution field is to transform the original brightness difference of the curved copper-aluminum hardware surface into a polarization energy weight distribution with clear physical meaning. Thus, even under complex optical environments such as strong light irradiation during cable inspection or multi-source interference, it can still stably extract the wave energy of the micro-dot code on the surface of the temperature measuring well, ensuring the anti-interference identification benchmark of the temperature measuring ceramic chip encapsulated within.

[0029] Step S13: Perform directional spatial filtering on the wave energy distribution field to generate a texture feature map, and perform feature calculation on the texture feature map to obtain quality assessment parameters.

[0030] To further refine the contour of the laser micro-dot code region, directional spatial filtering is performed on the wave-like energy distribution field using a phase angle distribution set. Specifically, the phase angle distribution set A reflects the digital spatial distribution of the vibration direction of the reflected light vector relative to the rotation angle of the X-axis. The reflected light vector refers to the composite physical vector characterizing the vibration state of the electric field intensity, formed by the superposition of specular and diffuse reflection light at the pixel location of the photosensitive chip. Since the edges of the micro-etched pits within the laser micro-dot code region cause a sharp phase angle jump in the diffuse reflection component of the reflected light vector at the microscale, a local spatial window is set. The size of the local spatial window is determined based on the ratio between the physical diameter of the micro-etched pits within the laser micro-dot code region and the single pixel size of the photosensitive chip of the imaging sensor. The purpose is to ensure that the local spatial window can completely cover the edge phase jump region of a single micro-etched pit, while avoiding the introduction of interference signals from adjacent micro-etched pits. An adaptive convolution operation guided by the phase angle distribution set is then performed on the wave-like energy distribution field. Specifically: For each pixel to be processed in the wave energy distribution field, a local spatial window with the pixel to be processed as its geometric center is extracted within the phase angle distribution set. The phase angle value corresponding to each pixel sampling point within the local spatial window is extracted. The algebraic difference between the phase angle value corresponding to each pixel sampling point within the local spatial window and the phase angle value corresponding to the pixel to be processed is calculated and defined as the phase gradient value. The phase gradient value is used to quantify the phase jump intensity caused by the edge of the micro-etched pits in the laser micro-dot code region. All phase gradient values ​​within the local spatial window are normalized to obtain a convolution kernel weight matrix corresponding to the size of the local spatial window. Each weight coefficient in the convolution kernel weight matrix is ​​positively correlated with the phase gradient value at the corresponding position. The convolution kernel weight matrix and the wave energy feature value within the corresponding local spatial window in the wave energy distribution field are arithmetically multiplied at the corresponding positions and then summed to obtain the filter response value corresponding to the pixel to be processed. The filtered response value of each pixel to be processed is compared with a preset response threshold. Pixels with filtered response values ​​greater than or equal to the preset threshold are extracted and defined as contour mapping points. The response threshold is set based on the numerical distribution range of the filtered response values ​​output after adaptive convolution of the curved copper-aluminum hardware. The maximum value of the filtered response value on the surface of the curved copper-aluminum hardware is taken as the discrimination boundary. The contour mapping points serve as digital coordinate pairs constituting the texture features of the laser micro-dot code region. Spatially, they exhibit anisotropic abrupt changes representing the geometric edges of the micro-etched pits; that is, the phase gradients of the contour mapping points in different directions have non-uniform intensity differences. An interpolation algorithm is used to connect the contour mapping points to generate a texture feature map representing the contour of the laser micro-dot code region.The texture feature map is a digital layer that eliminates background interference from curved copper and aluminum hardware surfaces and has high-contrast edges.

[0031] The quality assessment parameters are output synchronously. These parameters are obtained by extracting the spatial coordinates of contour mapping points in the texture feature map within the observation coordinate system. The spatial coordinates consist of X-axis coordinates, Y-axis coordinates, and a fixed zero-position Z-axis coordinate located on the surface of the photosensitive chip. Wave energy feature values ​​corresponding to these spatial coordinate positions are extracted from the wave energy distribution field to form a code point feature set; wave energy feature values ​​for all pixel positions other than those corresponding to the spatial coordinate positions are extracted from the wave energy distribution field to form a background feature set. The mean of the code point feature set and the standard deviation of the background feature set are calculated. An arithmetic division operation is then performed between these two sets to obtain the quality assessment parameters. These parameters are quantitative values ​​used to quantify the feature significance of contour features in the texture feature map relative to the background component in the wave energy distribution field.

[0032] Step S10 addresses the technical challenge of traditional imaging techniques when processing laser micro-dot codes on curved copper and aluminum hardware surfaces. This is due to severe semantic aliasing between the reflected light from the metal mirror and the code signal in a single grayscale dimension, leading to noise overwhelming the signal and preventing feature separation. Step S10 achieves physical-level energy decoupling of the temperature well reflection features, high-fidelity restoration of the micro-etched contours, and quantitative evaluation of feature acquisition quality. Specifically, the polarization feature set provides a multi-dimensional physical discrimination benchmark to distinguish between saturated noise generated by mirror reflection and texture in the laser micro-dot code region; the wave-like energy distribution field achieves nonlinear suppression of particle-like specular highlights at the pixel level; the texture feature map locks the anisotropic abrupt change features at the code point edges within the physical noise floor, outputting a high-contrast digital layer; and the quality evaluation parameters provide a closed-loop judgment index reflecting the saliency of features for the entire system's computation chain.

[0033] Step S20: Construct and apply an AI image-locking and grasping model based on quality assessment parameters to obtain target recognition confidence value; perform binarization judgment on target recognition confidence value to obtain geometric feature set; construct and apply an AI polymer image-locking model based on geometric feature set to obtain candidate code point set; perform spatial mapping on candidate code point set to obtain micro-point logic array.

[0034] Further, step S20 includes: Step S21: Construct and apply an AI image capture model based on quality assessment parameters to obtain target recognition confidence value. Perform binarization judgment based on target recognition confidence value to obtain geometric feature set.

[0035] After obtaining the texture feature map and quality assessment parameters, in order to capture the positioning drift caused by the attenuation of feature saliency due to surface wear on curved copper and aluminum fittings, a dynamic admission space for target recognition is defined within the observation coordinate system using the quality assessment parameters. During the processing of the texture feature map, an adaptive locking process is established that maps the quantized feedback of the quality assessment parameters to the decision parameters of the pre-built AI image-grabbing model.

[0036] The construction process of the AI ​​image capture model is as follows: First, acquire a training sample set. Collected laser-etched dot matrix images of curved copper and aluminum hardware surfaces with different radii of curvature are defined as positive sample images, and metal scratch images without laser-etched dot matrix images are defined as negative sample images. The training sample set consists of both positive and negative sample images. Second, build a neural network architecture. A traditional convolutional neural network model is used as the feature extraction backbone network. Based on the spatial arrangement of the micro-etched pits in the laser micro-dot code region, an ideal dot matrix topology model is established. The ideal dot matrix topology model is a mathematical reference matrix used to characterize the theoretical geometric center coordinate distribution of micro-etched pits in a standard plane. A dot matrix topology-aware operator is connected to the output of the final convolutional layer of the traditional convolutional neural network model. The dot matrix topology-aware operator is a mathematical operation unit used to calculate the spatial cross-correlation coefficient between the local pixel mapping matrix extracted by the final convolutional layer and the ideal dot matrix topology model. A pooling layer is connected to the output of the dot matrix topology-aware operator. The input of the multi-scale manifold attention mechanism is connected to the output of the pooling layer. The output of the multi-scale manifold attention mechanism is then connected to the input of the fully connected layer. The multi-scale manifold attention mechanism is a digital logical weighting unit that uses multiple filters of different sizes to perform significant weighting on feature channels based on the nonlinear geometric deformation caused by curved copper and aluminum fittings. Through the cascaded connections between these layers, a generative neural network architecture is constructed. The third step is model training. The training sample set is input into the neural network architecture, and parameter iteration is performed using a multi-task weighted joint loss function. The multi-task weighted joint loss function consists of a weighted sum of the FocalLoss classification loss term (used to quantify class determination error) and the CIoU regression loss term (used to quantify bounding box localization deviation). The weight parameters within the neural network architecture are adjusted using the error backpropagation algorithm until the average localization accuracy output by the neural network architecture reaches the target localization accuracy index. The target positioning accuracy index is set based on calculating the maximum logical offset displacement allowed by the laser micro-dot code when performing spatial topology reconstruction using the upper limit of the symbol fault tolerance of the Reed-Solomon error correction algorithm, and mapping the maximum logical offset displacement to the geometric constraint boundary for training convergence. This completes the construction of the AI ​​image capture model.

[0037] A recognition threshold weight is set based on quality assessment parameters. This threshold weight is determined by utilizing the dynamic range of reflected light intensity under different oxidation levels of curved copper-aluminum hardware. The empirical coefficient used to maintain a balance in recognition sensitivity is determined by calculating the probability density gradient difference between feature points and noise points in the texture feature map. The quality assessment parameters and the recognition threshold weight are then arithmetically multiplied to obtain the region of interest (ROI) recognition threshold. This ROI recognition threshold serves as a digital logical threshold for determining the presence of laser micro-dot code regions. After determining the ROI recognition threshold, an AI-based image-locking model is applied to perform target recognition on the texture feature map. Specifically, an adaptive sliding search window is set within the observation coordinate system, starting from the geometric center of the texture feature map. The adaptive sliding search window refers to a digital addressing operator that performs local feature traversal on the texture feature map. The size of the adaptive sliding search window is determined based on the mapping ratio between the physical side length of the laser micro-dot code region and the single-pixel physical resolution of the imaging sensor. This ensures the adaptive sliding search window completely encompasses the pixel matrix corresponding to a single laser micro-dot code region. The sliding step size of the adaptive sliding search window is determined based on the average arrangement period of adjacent micro-etched pits within the laser micro-dot code region. Brightness gradient information is extracted from the pixel field covered by the adaptive sliding search window. The rate of change of brightness along the X-axis and Y-axis of the observation coordinate system for each pixel sampling point within the adaptive sliding search window is calculated to obtain the pixel brightness gradient components. Spatial integration with respect to pixel coordinates is performed on all pixel brightness gradient components to obtain the local pixel second-order moment features. The local pixel second-order moment features refer to the numerical set representing the local texture structure tensor, which consists of the sum of the squares of the pixel brightness gradient components along the X-axis, the sum of the squares along the Y-axis, and the sum of their products along the X and Y axes for each pixel sampling point. The local pixel second-order moment features are input into the AI ​​image-locking and grasping model, outputting the target recognition confidence value. The target recognition confidence score refers to the probability of similarity that quantifies the topological features of the laser micro-dot code region matrix.

[0038] The target recognition confidence score is compared with the region of interest (ROI) recognition threshold. When the target recognition confidence score is greater than or equal to the ROI recognition threshold, the area covered by the current adaptive sliding search window is locked as the ROI. The ROI refers to a digital local field containing the complete laser micro-dot code area within the observation coordinate system. The center coordinates of the ROI within the observation coordinate system are extracted simultaneously. The center coordinates consist of the geometric center's coordinates on the X-axis, Y-axis, and the Z-axis coordinates corresponding to the fixed zero-position coordinates on the surface of the photosensitive chip. The local radius of curvature R corresponding to the ROI is also obtained. The local radius of curvature R is a physical parameter that quantifies the surface curvature of the copper-aluminum fitting at the location of the ROI. Using the center coordinates, phase angle values ​​within the pixel range of the ROI are extracted from the phase angle distribution set. The local radius of curvature R is calculated by performing a least-squares fitting operation using the microscopic geometric normal orientation information represented by the phase angle values. Considering that temperature measurement wells typically need to be laid in close proximity to high-voltage cables, their surfaces exhibit small-radius arc shapes of varying specifications. Extracting this local radius of curvature can provide a precise geometric compensation basis for subsequently eliminating deformation interference caused by special physical shapes. The region of interest, center coordinates, and local radius of curvature are digitally encapsulated and combined to obtain a geometric feature set.

[0039] Step S22: Construct and apply an AI polymer image-locked model based on the geometric feature set to obtain a set of candidate code points.

[0040] After obtaining the geometric feature set, in order to capture the edge adhesion effect of symbols caused by geometric nonlinearity in curved copper-aluminum fittings—that is, the pixel overlap and shape distortion features caused by curvature compression of micro-etched pits in the laser micro-dot code region in visual imaging—a dynamic mapping space for performing pixel-level stripping is defined within the observation coordinate system using the geometric feature set. This solves the problem of symbol centroid extraction deviation caused by traditional rectangular cutting algorithms when processing small-radius curved surfaces. During the processing of the region of interest, the micro-etched pits, as independent geometric operators constituting digital information, have their edge contours in the pixel plane strictly constrained by the nonlinear manifold contraction law determined by the local radius of curvature. Since the geometric feature set provides a high-fidelity spatial position reference and physical deformation parameters, a refined cutting process is established that maps the pixel distribution of micro-etched pits to a pre-constructed AI polymer image-locking model. The aim is to transform macroscopic region locking into an active symbol feature stripping process with physical topological constraints.

[0041] The AI ​​polymer image-locked model refers to a digital fine-cutting model that draws on the interaction logic of polymer chain segments and abstracts pixel topological relationships into energy-constrained monomer connection structures. The construction process is as follows: Boundary space information is extracted from the geometric feature set. This boundary space information refers to the set of coordinate values ​​used to define the closed perimeter of the region of interest within the observation coordinate system. Based on the boundary space information, the pixel coordinate range is determined within the observation coordinate system. This pixel coordinate range refers to the value interval of the discrete pixel row and column indices within the pixel grid of the photosensitive chip, defined by the boundary space information. The pixel row span and pixel column span values ​​corresponding to the region of interest are extracted based on the pixel coordinate range. A grid array, i.e., the dynamic mapping space, is constructed within the observation coordinate system, with the number of rows and columns aligned with the pixel row span and pixel column span values, respectively. See also... Figure 3 This is a schematic diagram of the relationship between the dynamic mapping space and the topological constraints of a single unit provided in an embodiment of the present invention. Figure 3 In the mesh structure shown, the dynamic mapping space consists of multiple mesh nodes arranged in pixel row and column order. The number of rows in the mesh array is aligned with the pixel row span, and the number of columns is aligned with the pixel column span. Each mesh node is a single cell. Figure 3 In the diagram, solid blue dots represent individual units, and dashed borders represent the boundary of the dynamic mapping space. Figure 3 It also demonstrates the mechanical constraint relationship between adjacent single units: the bidirectional arrows represent the topological constraint force generated by geometric shrinkage stress, which is used to constrain the degree of deviation of adjacent single units from the equilibrium distance and provide a topological self-recovery trend; the unidirectional arrows represent the centripetal traction force generated by gravitational weights, pointing towards the geometric center of the micro-etched pits, which is used to simulate the centripetal traction of the symbol center on the edge pixels. Figure 3 An example of a 4x5 grid array is provided, where the solid red dots represent the individual units to be processed, marked in the center of the grid. The individual unit to be processed is subjected to the combined mechanical action of the adjacent individual units within the eight-neighbor range. Through the synergistic constraint of geometric contraction stress and gravitational weight, pixel-level stripping of the sticky areas at the edge of the symbol is achieved, and finally a dynamic arrangement state representing the degree of aggregation of pixel clusters is generated.

[0042] The grid array contains multiple grid nodes arranged in pixel row and column order, with each grid node defined as a single unit. The single unit serves as the smallest dynamic calculation node carrying texture feature values. Texture feature values ​​refer to the digital values ​​extracted from the corresponding pixel coordinates in the texture feature map, representing the intensity of the etched edge. The equilibrium distance between single units is determined using the etching spacing of the laser micro-dot code region. The etching spacing refers to the physical distance between the geometric centers of two adjacent micro-etched pits on the substrate surface of the laser micro-dot code region, including horizontal and vertical etching spacing. The horizontal etching spacing refers to the distance between adjacent micro-etched pits arranged along the X-axis of the observation coordinate system, and the vertical etching spacing refers to the distance between adjacent micro-etched pits arranged along the Y-axis of the observation coordinate system. The equilibrium distance is the pixel displacement reference value obtained by mapping the etching spacing onto the pixel grid of the photosensitive chip, used to characterize the theoretical spatial positional relationship between adjacent single units. A linear mapping equation is established with the displacement distance between two adjacent single units as the independent variable and the geometric shrinkage stress as the dependent variable. The displacement distance refers to the real-time Euclidean distance between two adjacent unit cells in the coordinate plane defined by the X and Y axes of the observation coordinate system. Geometric shrinkage stress is used to quantify the degree to which adjacent unit cells deviate from their equilibrium distance and to provide a logical constraint component for the topological self-restoring trend. The formula for calculating geometric shrinkage stress is: Where Z represents the geometric shrinkage stress and L represents the displacement distance. The equilibrium distance is represented by K, which is a virtual elasticity coefficient used to adjust the spatial constraint stiffness between individual units. The setting is based on dynamically compensating for the pixel compression intensity between two adjacent micro-etched pits caused by the surface curvature of the curved copper-aluminum fittings using the local radius of curvature R. For example, the virtual elasticity coefficient is set to 1.5. A gravity weight calculation logic is designed, which refers to a digital mapping rule for quantifying the centripetal traction intensity experienced by an individual unit. Specifically, a mapping equation is established between the gradient component (independent variable: texture feature value) and the local radius of curvature R, and the gravity weight G (dependent variable). The gravity weight is used to simulate the centripetal traction force of the geometric center of the micro-etched pit on the edge pixels. The calculation formula is: ,in, The gradient component represents the gradient of the texture feature values ​​carried by a single unit. This gradient component is used to quantify the spatial rate of change of the texture feature values ​​of the unit to be processed relative to its neighboring units. In the observation coordinate system, the texture feature values ​​corresponding to the units adjacent to the unit to be processed along the X-axis are extracted, and the absolute value of the algebraic deviation between these values ​​is calculated, defined as the horizontal gradient value. Similarly, the texture feature values ​​corresponding to the units adjacent to the unit to be processed along the Y-axis are extracted, and the absolute value of the algebraic deviation between these values ​​is calculated, defined as the vertical gradient value. The squares of the horizontal and vertical gradient values ​​are summed to obtain the gradient component. The gravity weight adjusts the peeling intensity using the inverse relationship of the local radius of curvature. By increasing the gravity weight in pixel regions with greater surface curvature (i.e., smaller local radius of curvature), the edge contraction of the symbol caused by perspective compression is compensated. This completes the construction of the AI ​​polymer lock-in model.

[0043] An AI-powered polymer image-locked model is applied to perform symbol segmentation on the geometric feature set. Specifically, each texture feature value within the region of interest is filled into individual unit cells within the dynamic mapping space. The geometric contraction stress and gravitational weights of each unit cell due to deviation from its ideal position are calculated, and the gravitational weights of each unit cell in the X and Y axes of the observation coordinate system are extracted. The gravitational difference between adjacent unit cells in the X-axis direction is calculated to obtain the horizontal displacement component; the gravitational difference between adjacent unit cells in the Y-axis direction is calculated to obtain the vertical displacement component. The horizontal and vertical displacement components are vector-combined to obtain a displacement vector. This vector-combination operation transforms the scalar deviation between the two orthogonal directions of the horizontal and vertical displacement components into a geometric vector with direction and magnitude. For each individual element, the position coordinate difference between the individual element and its adjacent elements in the X and Y axes is calculated. The position coordinate difference is divided by the displacement distance between the individual element and its adjacent elements to obtain a unit direction vector. The geometric shrinkage stress is then arithmetically multiplied by the unit direction vector to obtain a correction vector. The spatial coordinates of each individual element in the observation coordinate system are linearly added to the corresponding displacement vector, and the corresponding correction vector is simultaneously superimposed to generate the coordinates after position offset, thus obtaining the dynamic arrangement state of each individual element. The dynamic arrangement state refers to the set of digital spatial positions representing the degree of pixel cluster aggregation formed under the combined action of centripetal gravity and topological constraint forces. The pixel spacing of the photosensitive chip and the equilibrium distance are used to set a clustering threshold to define the pixel aggregation boundary of a single micro-etched pit. In the dynamic arrangement state, a single unit is randomly selected as the starting seed point. The eight-neighbor range of the starting seed point within the dynamic mapping space is retrieved. The eight-neighbor range refers to the spatial set within the dynamic mapping space composed of a grid array, centered on the starting seed point and consisting of eight single units adjacent to the starting seed point in both the pixel row index direction and the pixel column index direction. All single units within the eight-neighbor range are identified as examination nodes. The Euclidean distance between the offset spatial coordinates of the starting seed point and the offset spatial coordinates of the examination nodes is calculated and defined as the spatial distance. The spatial distance is compared with a clustering threshold. If the spatial distance is less than or equal to the clustering threshold, the starting seed point and the examination node are assigned the same attribute identifier. The attribute identifier is a unique coded label used to distinguish the ownership of different code elements in the digitization matrix. By recursively retrieving and labeling all single units with the same attribute identifier within the dynamic arrangement state, multiple independent physically connected components are generated, defined as candidate code points. All candidate code points are summarized to obtain a candidate code point set, which represents the digitization description set of each micro-etched unit in the laser micro-dot code region under the current distortion state. Each candidate code point in the candidate code point set contains the coordinates of the code point centroid and the code point feature value in the observation coordinate system.The barycentric coordinates of a code element refer to the combination of digitized coordinates obtained by averaging the offset spatial coordinates of individual units within a physically connected domain, used to characterize the center position of the candidate code point. Specifically: the offset spatial coordinates of each individual unit within the physically connected domain are extracted, and the arithmetic mean of the coordinate values ​​of all individual units in the X-axis direction, the Y-axis direction, and the Z-axis direction are calculated respectively; the arithmetic mean of the coordinate values ​​in these three dimensions constitutes the barycentric coordinates of the candidate code point. The symbol feature value refers to the total number of individual units contained in each physically connected domain. Even when extreme perspective distortion occurs on the surface of the temperature sensing well due to excessive curvature, it can still ensure that the microscopic etched pits representing the identity of the temperature sensing ceramic chip are clearly and independently segmented, fundamentally avoiding misjudgment of node identity caused by feature adhesion.

[0044] Step S23: Perform spatial transformation operation on the candidate code point set to generate reconstructed coordinate pairs, and map the reconstructed coordinate pairs to obtain the micro-point logic array.

[0045] After obtaining the candidate code point set, in order to capture the symbol topological misalignment effect caused by geometric nonlinearity of curved copper-aluminum fittings—that is, the nonlinear stretching and spatial position offset of each candidate code point in the candidate code point set due to surface projection in visual imaging—a geometric transformation space for performing manifold restoration is defined within the observation coordinate system using a geometric feature set. The aim is to transform the passive, distorted pixel position distribution into an active, row-column aligned micro-point logic array, thus solving the problem of symbol spatial topological distortion caused by curvature interference.

[0046] Specifically, a reconstruction weight distribution set is constructed. Local radii of curvature and center coordinates are extracted from the geometric feature set. For each candidate code point in the candidate code point set, the corresponding barycentric coordinates of the code element are extracted. The displacement deviation of the barycentric coordinates of each candidate code point relative to the center coordinates in the X-axis direction is calculated. The azimuth rotation angle is obtained by calculating the arcsine function value of the ratio of the displacement deviation value to the local radius of curvature. The azimuth rotation angle represents the angular displacement of the candidate code point's physical position on the three-dimensional curved surface relative to the central axis of the observation coordinate system. A focal length compensation coefficient is introduced. The focal length compensation coefficient is a digital correction factor determined based on the focal length of the optical lens of the imaging sensor, used to correct the scale scaling deviation caused by perspective projection; for example, it is set to 1.5. Based on the local radius of curvature, the azimuth rotation angle, and the focal length compensation coefficient, a horizontal reconstruction weight C1 and a vertical reconstruction weight C2 are calculated for each candidate code point. The horizontal reconstruction weight is used as a digital horizontal compensation operator to restore the linear projection distance of the curved copper-aluminum hardware surface to the physically true arc length. The calculation formula is: ,in, , and These represent the focal length compensation coefficient, azimuth rotation angle, and displacement deviation value, respectively. The vertical reconstruction weight is used to correct the digital vertical compensation operator for longitudinal scale reduction caused by the deviation of the surface normal vector of the curved copper-aluminum fitting from the imaging optical axis. The calculation formula is as follows: Where cos(·) is the cosine function. The horizontal and vertical reconstruction weights corresponding to each candidate code point together constitute the reconstruction weight distribution set. The X-axis and Y-axis coordinates of the extracted code point centroid coordinates are arithmetically multiplied with the horizontal reconstruction weights and X-axis coordinates of the candidate code point to obtain the horizontal coordinate component; the vertical coordinate component is obtained by arithmetically multiplying with the vertical reconstruction weights and Y-axis coordinates; the horizontal and vertical components together constitute the reconstruction coordinate pair corresponding to the candidate code point.

[0047] Obtain the encoding scale parameters characterizing the original physical structure attributes of the laser micro-dot code region. The encoding scale parameters are digital values ​​determined by the etching process, used to quantify the total number of micro-etched pits in the horizontal and vertical directions within the laser micro-dot code region. Based on the total number in the horizontal direction in the encoding scale parameters, determine the column index scale of the micro-dot logic array; based on the total number in the vertical direction, determine the row index scale of the micro-dot logic array. The micro-dot logic array is a digital data structure composed of multiple binary logic levels arranged in row and column index order. Extract the preset etching spacing within the laser micro-dot code region. Obtain the single-pixel size value corresponding to the photosensitive chip in the imaging sensor. The single-pixel size value refers to the length specification occupied by each pixel unit on the photosensitive chip in physical space. Calculate the arithmetic ratio between the horizontal etching spacing and the single-pixel size value to obtain the horizontal pixel displacement step size, and determine the horizontal pixel displacement step size as the column spacing reference value; calculate the arithmetic ratio between the vertical etching spacing and the single-pixel size value to obtain the vertical pixel displacement step size, and determine the vertical pixel displacement step size as the row spacing reference value.

[0048] Within the observation coordinate system, the plane whose Z-axis coordinate, determined by the photosensitive surface of the photosensitive chip, is constantly zero is defined as the two-dimensional pixel plane. The center coordinates are extracted from the geometric feature set. Based on the geometric center position determined by the center coordinates within the two-dimensional pixel plane, combined with the column index size, row index size, and column and row spacing reference values, the grid logic origin is calculated. The grid logic origin refers to the digitized coordinate point corresponding to the first row and first column index of the micro-dot logic array within the two-dimensional pixel plane. The grid logic origin is determined as follows: using the center coordinates in the X-axis direction as a reference, subtracting half of the product of the column index size and column spacing reference value yields the horizontal offset of the grid logic origin; using the center coordinates in the Y-axis direction as a reference, subtracting half of the product of the row index size and row spacing reference value yields the vertical offset of the grid logic origin. Using the grid logic origin as the starting point, multiple coordinate points distributed in a rectangular array are generated within the two-dimensional pixel plane according to the column and row spacing reference values, and these are defined as logic nodes. The spatial topology framework composed of all logical nodes is the digital standard grid.

[0049] For each reconstructed coordinate pair, calculate the Euclidean distance between the reconstructed coordinate pair and each logical node within the digitized standard grid. Assign the reconstructed coordinate pair to the logical node with the smallest Euclidean distance. If a specific logical node in the digitized standard grid is assigned a corresponding reconstructed coordinate pair, then the array element value corresponding to that specific logical node is set to binary logic level 1; if a specific logical node in the digitized standard grid is not assigned a reconstructed coordinate pair, then the array element value corresponding to that specific logical node is set to binary logic level 0. By summing the array element values ​​corresponding to all logical nodes within the digitized standard grid, a micro-dot logic array is generated.

[0050] Step S20 solves the technical challenges of positioning drift caused by surface oxidation and wear in complex service environments for curved copper-aluminum fittings, as well as topological misalignment caused by symbol edge adhesion and manifold distortion due to small radius curvature, by constructing a geometric feature set, generating a candidate code point set, and reconstructing a micro-dot logic array. This achieves precise code region locking under interference, pixel-level independent symbol stripping, and geometric restoration from nonlinear manifolds to standard logic matrices. Specifically, the geometric feature set integrates an AI image-locking model to provide spatial constraints and physical deformation parameters for subsequent refined cutting; the candidate code point set, through the application of an AI polymer image-locking model, achieves precise transformation from fuzzy textures to independent logic units; and the micro-dot logic array maps the reconstructed coordinate points to a digital standard grid, generating a standardized binary logic matrix with row and column alignment attributes.

[0051] Step S30: Determine the logic node to be detected based on the micro-dot logic array, perform spatial topology extraction on the logic node to be detected to obtain a topology error correction map, perform feature compensation on the topology error correction map to obtain a logic verification pass identifier.

[0052] Further, step S30 includes: Step S31: Determine the logic node to be detected based on the micro-dot logic array, perform spatial topology extraction on the logic node to be detected, and obtain the topology error correction map.

[0053] After obtaining the micro-point logic array, in order to capture the feature defects caused by surface grid flashover, chemical corrosion, or foreign object obstruction in curved copper-aluminum fittings—that is, the feature dimension collapse caused by the partial binary logic level in the micro-point logic array changing from 1 to 0 due to physical damage—a probabilistic search space for performing logic repair is defined within the digital standard grid using the micro-point logic array. Considering that temperature measurement wells are easily damaged by external friction and natural corrosion during deep-buried cable trenches or long-term outdoor operation, the aim is to transform static logic judgment into an active probabilistic search domain extraction with topological self-healing capabilities, thereby maximizing the preservation of damaged temperature measurement well identification information and ensuring the continuity of cable temperature measurement data.

[0054] Specifically, the micro-dot logic array is traversed, and each logic node with a binary logic level of 0 is extracted and identified as the logic node to be tested. The logic node to be tested refers to the digitized position in the micro-dot logic array that requires topological consistency verification due to missing binary logic levels. For each logic node to be tested, its eight-neighbor range is searched within the digitized standard grid; all logic nodes with a binary logic level of 1 within the eight-neighbor range are extracted, and these extracted logic nodes are digitized to obtain an auxiliary reference point set. The auxiliary reference point set provides a local spatial topological reference for logical inference of missing features. For each logic node with a binary logic level of 1 in the auxiliary reference point set, the symbol centroid coordinates of that logic node in the observation coordinate system are extracted, and the corresponding spatial coordinates of the logic node to be tested are extracted simultaneously. The arithmetic square root of the sum of the squares of the coordinate deviations of the symbol centroid coordinates relative to the spatial coordinates in the X, Y, and Z axes is calculated to obtain the spatial geometric distance. The difference between the barycenter coordinates and spatial coordinates of the symbol in the two-dimensional pixel plane is calculated, and the arctangent function value of the ratio of the vertical axis value to the horizontal axis value in the coordinate difference is determined as the phase offset. The phase offset is used to quantify the spatial orientation of the logical nodes in the auxiliary reference point set relative to the logical node to be detected. The distance distribution probability corresponding to each logical node to be detected is determined based on the horizontal and vertical etching spacing. The distance distribution probability is used to quantify the digital value of the spatial association strength of the logical node to be detected relative to the auxiliary reference point set in the physical spacing dimension. The logic for determining the distance distribution probability is as follows: calculate the absolute value of the difference between the spatial geometric distance and the horizontal and vertical etching spacing, respectively, to obtain the absolute value of the distance deviation. Since the laser micro-dot code region strictly follows the periodic arrangement rule defined by the horizontal and vertical etching spacing during construction, any valid symbol originating from physical etching must maintain the above spacing with adjacent symbols in spatial distribution. The distance distribution probability is negatively correlated with the absolute value of the distance deviation; when the absolute value of the distance deviation approaches 0, the determined distance distribution probability reaches its maximum value. Specifically, a scale smoothing operator is set, the value of which is determined based on the ratio of the arithmetic mean of the horizontal etching spacing to the vertical etching spacing, for example, in the range of 1 to 2; the quotient between the square of the absolute value of the distance deviation and twice the square of the scale smoothing operator is calculated to obtain a first quotient value, and the opposite of the first quotient value is subjected to an exponential operation with the natural constant e as the base to obtain the distance distribution probability.

[0055] Based on the orthogonal arrangement characteristics of the digital standard grid, the directional distribution probability corresponding to each logical node to be detected is determined. Since the digital standard grid is a rectangular orthogonal frame composed of multiple logical nodes arranged in rows and columns, the relative azimuth angles of code elements belonging to the same encoding sequence in the two-dimensional pixel plane must be in the normal position of the horizontal or vertical axis. Therefore, using the axial direction of the digital standard grid as a reference, four mutually orthogonal standard pointing angles are determined, namely 0 degrees, 90 degrees, 180 degrees, and 270 degrees. The absolute values ​​of the differences between the phase deviation angle and the four standard pointing angles are calculated, and the minimum value among the obtained absolute values ​​of multiple differences is determined as the absolute value of the directional deviation. The directional distribution probability and the absolute value of the directional deviation have a negative correlation; when the absolute value of the directional deviation approaches 0, the determined directional distribution probability reaches its maximum value. Specifically, the cosine value of the absolute value of the directional deviation is taken, and this cosine value is multiplied by the shape feature factor to obtain the directional distribution probability. The shape feature factor is set according to the number of standard pointing angles; for example, it is set to 4.

[0056] The distance distribution probability and the direction distribution probability are linearly weighted and superimposed to obtain the logical recovery probability value corresponding to the logical node to be detected. The logical recovery probability value is a digital indicator used to quantify the degree of probability of feature dimension collapse of the logical node to be detected. The spatial set composed of the logical recovery probability values ​​corresponding to all logical nodes to be detected is determined as the topological error correction map. The role of the topological error correction map is to provide a digital discrimination basis with both physical and geometric dimensions for the logical completion of missing code elements.

[0057] Step S32: Perform feature compensation based on the topology error correction map to obtain a full-dimensional array, and apply ring redundancy check operation to the full-dimensional array to obtain a logical check pass flag.

[0058] After obtaining the topological error correction map, in order to recover the complete digital information originally carried by the laser micro-dot code region and solve the problem of local logic loss caused by physical damage to the metal surface, a probabilistic search domain is determined within the digital standard grid. The aim is to provide specific indexes of the sites to be repaired for the logical completion of missing code elements.

[0059] Calculate the feature dimension compensation factor for each logic node to be detected. This feature dimension compensation factor characterizes the decision parameter for the logic node to perform forced logic level correction. Specifically, obtain the probability response value of the logic node to be detected in the topology correction map. This probability response value refers to the logic recovery probability value corresponding to each logic node to be detected in the topology correction map. Using the spatial coordinates of the logic node to be detected, perform pixel-level indexing within the topology correction map to extract the digitized probability value of the corresponding position. Determine confidence weight coefficients for the logic nodes in the auxiliary reference point set. These confidence weight coefficients are weighted based on the edge sharpness of each logic node with a binary logic level value of 1, to reduce the risk of misinference in severely damaged areas. Specifically, in the texture feature map, a local pixel window of a preset size is extracted, centered on the spatial coordinates of each logic node with a binary logic level of 1 in the observation coordinate system. The size of the local pixel window is determined based on the ratio between the physical diameter of the micro-etched pits in the laser micro-dot code area and the physical size of a single pixel of the imaging sensor. This ensures that the local pixel window can completely enclose the edge phase transition region of a single micro-etched pit while avoiding the introduction of interference signals from adjacent micro-etched pits. In this embodiment, for example, the size of the local pixel window is set to 3×3 pixels. The luminance gradient magnitude of each pixel in the local pixel window is calculated, and the arithmetic mean of the luminance gradient magnitudes of all pixels in the local pixel window is defined as the edge sharpness value of the logic node. Based on the obtained quality evaluation parameters, the global gradient average is obtained by extracting the luminance gradient magnitude at each contour mapping point in the texture feature map and performing arithmetic accumulation, then dividing by the total number of contour mapping points. The confidence weight coefficient is obtained by calculating the ratio of the edge sharpness value to the sharpness benchmark.

[0060] Each probability response value is arithmetically multiplied with its corresponding confidence weight coefficient to obtain a series of weighted probability values. These weighted probability values ​​are then summed to obtain the feature dimension compensation factor. A recombination threshold is set, which is a digital probability truncation value used to determine whether a logic node under test has experienced feature dimension collapse. This threshold is set using the theoretical error correction redundancy of the laser micro-dot code region and the average oxidation wear ratio of the curved copper-aluminum hardware surface to establish a confidence boundary for logic repair; for example, it is set to 0.85. The feature dimension compensation factor corresponding to each logic node under test is compared with the recombination threshold. If the feature dimension compensation factor is greater than or equal to the recombination threshold, the logic node under test is determined to have experienced feature dimension collapse, and its binary logic level is forcibly corrected to 1. If the feature dimension compensation factor is less than the recombination threshold, the binary logic level of the logic node under test remains unchanged at 0. Based on the comparison and judgment results of each logic node to be detected, numerical update and state alignment operations are performed on each corresponding binary logic level in the micro-dot logic array to obtain a full-dimensional array. A full-dimensional array refers to a logic array that, after topological feature completion, completely represents the original encoding state of the laser micro-dot code region in its digital spatial topology. An extraction scanning path is set, which is the addressing trajectory within the digital standard grid used to convert two-dimensional distributed logic levels into one-dimensional timing signals. A reading sequence consistent with the encoding protocol corresponding to the laser micro-dot code region is established using the row and column index order in the encoding scale parameter. Each binary logic level is sequentially extracted from the full-dimensional array according to the extraction scanning path to construct a bitstream to be verified. Cyclic redundancy check (CRC) technology, such as the CRC-16 check algorithm, is used to perform CRC operation on the bitstream to be verified. If the remainder of the CRC operation is zero, it is determined that the full-dimensional array, after topological feature completion, completely restores the original encoding information of the laser micro-dot code region, generating a logic verification pass flag, which is a logic trigger signal used to characterize the physical consistency of the end-to-end data processing results. By verifying the identifier through logical checks, the output original encoded information is confirmed to be a valid identifier of the temperature-measuring ceramic chip, thereby achieving reliable identification of the cable temperature-measuring well node.

[0061] Step S30 addresses the technical challenges of local logic loss due to surface physical damage in curved copper-aluminum fittings and decoding interruption caused by the failure of the error correction mechanism due to missing underlying features by constructing a topology error correction map, generating a full-dimensional array, and establishing a logic verification pass marker. This achieves topology self-repair of incomplete features, complete data logic completion throughout the entire lifecycle, and closed-loop verification of decoding result consistency. Specifically, the topology error correction map integrates distance and direction distribution probabilities, providing a priori probability basis in both physical and geometric dimensions for the logic recovery of the detected logic node. The full-dimensional array performs intelligent correction of the logic level from 0 to 1 by performing feature dimension compensation based on correlation probability, restoring the original encoding state of the laser micro-dot code area. The logic verification pass marker applies cyclic redundancy check (CRC) operations to perform closed-loop verification on the completed bitstream, ensuring that the end-to-end solution results are highly faithful to the physical entity information, thus achieving the technical goal of reliable identification of cable temperature measurement nodes in harsh industrial environments.

[0062] Example 2 This embodiment, based on Embodiment 1, provides a noise-resistant temperature-measuring laser micro-dot code decoding system for curved copper and aluminum fittings, such as... Figure 4 As shown, it includes: Energy decoupling module: used to acquire the original image set of the curved copper and aluminum hardware surface, perform polarization analysis operation on the original image set to obtain the polarization feature set, perform energy decoupling on the polarization feature set to obtain the wave energy distribution field, perform directional spatial filtering on the wave energy distribution field to generate a texture feature map, and perform feature calculation on the texture feature map to obtain quality assessment parameters; Manifold Reconstruction Module: This module is used to construct and apply an AI-based image-locking and grasping model based on quality assessment parameters, obtain target recognition confidence values, perform binarization judgment on the target recognition confidence values ​​to obtain a geometric feature set, construct and apply an AI polymer image-locking model based on the geometric feature set to obtain a candidate code point set, and perform spatial mapping on the candidate code point set to obtain a micro-point logic array. Topology correction module: used to determine the logic node to be detected based on the micro-point logic array, perform spatial topology extraction on the logic node to be detected to obtain the topology correction map, perform feature compensation on the topology correction map to obtain the logic verification pass mark.

[0063] In the energy decoupling module, the process involves acquiring the original image set of the curved copper-aluminum hardware surface, performing polarization analysis on the original image set to obtain a polarization feature set, performing energy decoupling on the polarization feature set to obtain a wave-like energy distribution field, performing directional spatial filtering on the wave-like energy distribution field to generate a texture feature map, and performing feature calculation on the texture feature map to obtain quality assessment parameters, including: Step S11: Obtain the original image set of the curved copper-aluminum hardware surface and construct the observation coordinate system. Perform polarization analysis operation on the original image set according to the observation coordinate system to obtain the polarization feature set. Step S12: Perform energy decoupling using the polarization feature set to obtain the wave-like energy distribution field; Step S13: Perform directional spatial filtering on the wave energy distribution field to generate a texture feature map, and perform feature calculation on the texture feature map to obtain quality assessment parameters.

[0064] In the manifold reconstruction module, the step of constructing and applying an AI-based image-locking model based on quality assessment parameters to obtain a target recognition confidence value, performing binarization on the target recognition confidence value to obtain a geometric feature set, constructing and applying an AI polymer image-locking model based on the geometric feature set to obtain a candidate code point set, and performing spatial mapping on the candidate code point set to obtain a micro-point logic array, includes: Step S21: Construct and apply an AI image capture model based on quality assessment parameters to obtain target recognition confidence value; perform binarization judgment based on target recognition confidence value to obtain geometric feature set. Step S22: Construct and apply an AI polymer image-locking model based on the geometric feature set to obtain a candidate code point set; Step S23: Perform spatial transformation operation on the candidate code point set to generate reconstructed coordinate pairs, and map the reconstructed coordinate pairs to obtain the micro-point logic array.

[0065] In the topology correction module, the process of determining the logic node to be detected based on the micro-point logic array, performing spatial topology extraction on the logic node to be detected to obtain a topology correction map, performing feature compensation on the topology correction map, and obtaining a logic verification pass identifier includes: Step S31: Determine the logic node to be detected based on the micro-dot logic array, perform spatial topology extraction on the logic node to be detected, and obtain the topology error correction map; Step S32: Perform feature compensation based on the topology error correction map to obtain a full-dimensional array, and apply ring redundancy check operation to the full-dimensional array to obtain a logical check pass flag.

[0066] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0067] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A noise-resistant temperature-sensing laser micro-dot code decoding method for curved copper and aluminum fittings, characterized in that, The method includes: The original image set of the curved copper-aluminum hardware surface is obtained. Polarization analysis is performed on the original image set to obtain a polarization feature set. Energy decoupling is performed on the polarization feature set to obtain a wave energy distribution field. Directional spatial filtering is performed on the wave energy distribution field to generate a texture feature map. Feature calculation is performed on the texture feature map to obtain quality assessment parameters. Based on the quality assessment parameters, an AI-based image capture model is constructed and applied to obtain the target recognition confidence value. The target recognition confidence value is then binarized to obtain a geometric feature set. Based on the geometric feature set, an AI-based polymer image capture model is constructed and applied to obtain a candidate code point set. The candidate code point set is then spatially mapped to obtain a micro-dot logic array. The logic nodes to be detected are determined based on the micro-point logic array. Spatial topology extraction is performed on the logic nodes to be detected to obtain the topology error correction map. Feature compensation is performed on the topology error correction map to obtain the logic verification pass mark.

2. The noise-resistant temperature measurement laser micro-dot code decoding method for curved copper and aluminum fittings as described in claim 1, characterized in that, The method for obtaining the polarization feature set includes: The imaging sensor is used to collect information on the intensity of reflected light with different polarization angles, and a two-dimensional brightness distribution matrix consisting of multiple pixel sampling points is generated, which is the original image set. An observation coordinate system is constructed based on the original image set. For each pixel sampling point in the original image set, a parameter distribution set is parsed out. The parameter distribution set includes a total light intensity parameter representing macroscopic brightness, a horizontal linear polarization parameter representing the difference between horizontal and vertical vibration, and a 45-degree linear polarization parameter representing the difference between oblique vibration. Physical parameter mapping is performed based on the parameter distribution set to generate a polarization degree distribution set that characterizes the proportion of linear polarization energy, and a phase angle distribution set that characterizes the rotation angle of the reflected light vector vibration direction relative to the observation coordinate system. The polarization degree distribution set and the phase angle distribution set are then combined and encapsulated into a polarization feature set.

3. The noise-resistant temperature-measuring laser micro-dot code decoding method for curved copper-aluminum fittings as described in claim 2, characterized in that, The wave-like energy distribution field includes: A wave-particle duality adjustment factor is introduced, and the energy weight corresponding to each pixel sampling point is calculated by using the numerical values ​​of the polarization degree distribution set and the wave-particle duality adjustment factor. The total light intensity parameter in the parameter distribution set is multiplied by the energy weight at the corresponding position using pixel-level arithmetic multiplication. The energy intensity of the wave component dominated by diffuse reflection in the reflected light field is extracted to obtain the wave energy characteristic value. The wave energy characteristic values ​​corresponding to all pixel sampling points on the photosensitive chip are mapped to a two-dimensional digital set, which is defined as the wave energy distribution field.

4. The noise-resistant temperature measurement laser micro-dot code decoding method for curved copper-aluminum fittings as described in claim 3, characterized in that, The methods for obtaining the quality assessment parameters include: The phase gradient values ​​within a preset local spatial window are calculated using the phase angle distribution set, and a convolution kernel weight matrix is ​​generated. An adaptive convolution operation is performed on the wave energy distribution field using the convolution kernel weight matrix to obtain the filtered response value; Pixels whose filtered response values ​​are greater than or equal to a preset response threshold are extracted as contour mapping points, and texture feature maps are generated by connecting the contour mapping points. Obtain the spatial coordinates of the contour mapping points in the observation coordinate system, extract the values ​​of the corresponding spatial coordinate positions from the wave energy distribution field to form a code point feature set, and extract the values ​​of pixel positions other than the corresponding spatial coordinate positions to form a background feature set. Calculate the mean of the code point feature set and the standard deviation of the background feature set, and then perform an arithmetic division operation between the two to obtain the quality assessment parameters.

5. The noise-resistant temperature-measuring laser micro-dot code decoding method for curved copper-aluminum fittings as described in claim 4, characterized in that, The method for obtaining the geometric feature set includes: Set a recognition threshold weight, perform an arithmetic product operation between the quality assessment parameter and the recognition threshold weight, and obtain the interest region recognition threshold. Build and apply an AI-based image capture model to perform a sliding search on the texture feature map and output the target recognition confidence score. When the target recognition confidence score is greater than or equal to the region of interest recognition threshold, the region of interest is locked. Obtain the center coordinates of the region of interest in the observation coordinate system, and use the phase angle distribution set to fit and calculate the local radius of curvature corresponding to the region of interest; The region of interest, center coordinates, and local radius of curvature are digitally encapsulated to obtain a geometric feature set.

6. The noise-resistant temperature-measuring laser micro-dot code decoding method for curved copper-aluminum fittings as described in claim 5, characterized in that, The method for obtaining the candidate code point set includes: Based on the geometric feature set, a grid array corresponding to the pixel distribution of the region of interest is constructed as a dynamic mapping space. Each node in the grid array is defined as a single unit, and each single unit is filled with texture feature values. The balance distance between individual units is determined by using a preset etching spacing, and the gradient components of the texture feature values ​​are calculated. Calculate the geometric contraction stress and gravitational weight of a single unit cell; whereby the geometric contraction stress is determined based on the degree to which the real-time displacement distance of adjacent single units deviates from the equilibrium distance, and the gravitational weight is determined based on the gradient component and the local radius of curvature. Based on the vector synthesis result of geometric contraction stress and gravitational weight, the single unit is driven to shift its position in the observation coordinate system, generating a dynamic arrangement state; By recursively retrieving and labeling all individual units within the dynamic arrangement state, multiple mutually independent physically connected domains are generated, which are defined as candidate code points. All candidate code points are then aggregated to obtain a set of candidate code points, which includes the barycenter coordinates of the code points in the observation coordinate system and the code point feature values.

7. The noise-resistant temperature-measuring laser micro-dot code decoding method for curved copper-aluminum fittings as described in claim 6, characterized in that, The method for obtaining the micro-dot logic array includes: Set the focal length compensation coefficient, and calculate the reconstruction weight distribution set for each candidate code point in the candidate code point set based on the local radius of curvature, center coordinates and focal length compensation coefficient. The reconstruction weight distribution set includes horizontal reconstruction weight and vertical reconstruction weight. The code centroid coordinates are weighted using horizontal and vertical reconstruction weights to generate reconstruction coordinate pairs. Obtain the coding scale parameter that characterizes the original physical structure properties of the laser micro-dot code region, and construct a digital standard grid in the observation coordinate system based on the coding scale parameter and the etching spacing. The digital standard grid contains multiple logical nodes. Calculate the Euclidean distance between the reconstructed coordinate pair and the logical node, and assign the reconstructed coordinate pair to the logical node with the smallest Euclidean distance; The binary logic level is determined based on the allocation status of the logic nodes, and the binary logic levels of all logic nodes are combined in the row and column order of the rectangular array to obtain the micro-dot logic array.

8. The noise-resistant temperature-measuring laser micro-dot code decoding method for curved copper-aluminum fittings as described in claim 7, characterized in that, The topology error correction map includes: Extract nodes in the micro-dot logic array whose binary logic level is zero, and define them as logic nodes to be detected; Search the eight-neighbor range of the logic node to be detected within the digital standard grid, and extract the nodes with a binary logic level of one to form an auxiliary reference point set; The distance distribution probability of the logic node to be detected relative to the auxiliary reference point set is calculated based on the etching spacing, and the direction distribution probability of the logic node to be detected is calculated based on the orthogonal arrangement characteristics of the digital standard grid. The distance distribution probability and the direction distribution probability are linearly weighted and superimposed to obtain the logical recovery probability value. The logical recovery probability values ​​corresponding to all logical nodes to be detected constitute the topology error correction map.

9. The noise-resistant temperature-measuring laser micro-dot code decoding method for curved copper-aluminum fittings as described in claim 8, characterized in that, The logical verification pass identifier includes: To assist in the distribution of node confidence weight coefficients in the reference point set, the confidence weight coefficients are used to perform a weighted operation on the logical recovery probability values ​​in the topology error correction graph to obtain the feature dimension compensation factor of the logical node to be detected. The feature dimension compensation factor is compared with the preset recombination threshold. If the feature dimension compensation factor is greater than or equal to the recombination threshold, the binary logic level of the logic node to be detected is corrected to one to obtain a full-dimensional array. Set an extraction scan path, construct the bit stream to be verified from the full-dimensional array according to the extraction scan path, and perform cyclic redundancy check operation on the bit stream to be verified. If the remainder of the cyclic redundancy check operation is zero, a logical check pass flag is obtained.

10. A noise-resistant temperature-sensing laser micro-dot code decoding system for curved copper and aluminum fittings, used to implement the noise-resistant temperature-sensing laser micro-dot code decoding method for curved copper and aluminum fittings as described in any one of claims 1-9, characterized in that, The system includes: Energy decoupling module: used to acquire the original image set of the curved copper and aluminum hardware surface, perform polarization analysis operation on the original image set to obtain the polarization feature set, perform energy decoupling on the polarization feature set to obtain the wave energy distribution field, perform directional spatial filtering on the wave energy distribution field to generate a texture feature map, and perform feature calculation on the texture feature map to obtain quality assessment parameters; Manifold Reconstruction Module: This module is used to construct and apply an AI-based image-locking and grasping model based on quality assessment parameters, obtain target recognition confidence values, perform binarization judgment on the target recognition confidence values ​​to obtain a geometric feature set, construct and apply an AI polymer image-locking model based on the geometric feature set to obtain a candidate code point set, and perform spatial mapping on the candidate code point set to obtain a micro-point logic array. Topology correction module: used to determine the logic node to be detected based on the micro-point logic array, perform spatial topology extraction on the logic node to be detected to obtain the topology correction map, perform feature compensation on the topology correction map to obtain the logic verification pass mark.

Images