Ultraviolet imaging light path fusion calibration system based on corona discharge of power equipment

Abstract

The present application provides a power equipment corona discharge ultraviolet imaging light path fusion calibration system, relates to the technical field of corona discharge ultraviolet detection, and comprises an ultraviolet dot array emission module, a space reference coordinate module, a fusion deviation detection module, an axial error analysis module, a fusion compensation module and an automatic calibration control module. The system constructs a standard light reference source in a predetermined space through the ultraviolet dot array emission module, establishes a unified geometric mapping relationship between ultraviolet and visible light sensors by using the space reference coordinate module, extracts double-mode image features in real time by the fusion deviation detection module, quantifies the optical axis offset by the axial error analysis module, and compensates the imaging distortion in real time by the fusion compensation module through a dynamic calibration algorithm. Finally, the automatic calibration control module realizes closed-loop feedback, realizes adaptive correction of dynamic light path offset in the inspection process, effectively solves the problems of poor multi-sensor registration accuracy, large background interference and lack of real-time calibration feedback under complex power working conditions.

Description

Technical Field

[0001] This invention relates to the field of corona discharge ultraviolet detection technology, specifically to a corona discharge ultraviolet imaging optical path fusion calibration system for power equipment. Background Technology

[0002] With the continuous deepening of smart grid construction, the safe operation of high-voltage power equipment has become a core issue in ensuring energy supply. During operation, power equipment often experiences local corona discharge due to insulation defects, surface roughness, and sharp corner effects, leading to power loss, radio interference, and even serious insulation breakdown accidents. Currently, non-contact detection technology based on solar-blind ultraviolet imaging has become the mainstream method in the field of power inspection. This technology captures the ultraviolet photon stream generated by the discharge to achieve accurate location of the discharge and quantitative assessment of its severity. In order to achieve intuitive correspondence between ultraviolet discharge information and physical images of power equipment, dual-spectral optical path fusion (superposition of ultraviolet and visible light images) technology has been widely used, becoming a key basis for maintenance personnel to judge the nature and location of hidden dangers.

[0003] However, existing optical path fusion calibration technology for corona discharge ultraviolet imaging of power equipment still faces the following key technical challenges under complex actual operating conditions: 1. Multimodal image registration accuracy and spatial geometric distortion are difficult to coordinate. Due to the spatial offset (parallax) of the physical optical axis between ultraviolet imaging detectors and visible light cameras, and their significantly different imaging mechanisms, ultraviolet images typically present as sparse single-photon counting lattices, lacking natural common feature matching points with visible light images. Existing registration methods mostly rely on preset fixed parameter models, making it difficult to compensate in real time for nonlinear geometric distortions caused by changes in shooting distance and equipment posture jitter. This results in a significant shift in the mapping position of the discharge area on the visible light image, seriously misleading maintenance personnel in determining the fault location.

[0004] 2. Poor consistency between background light suppression and optical path calibration in complex environments. The operating environment of power equipment is complex. Strong sunlight, highly reflective backgrounds, and variable weather conditions can severely interfere with ultraviolet imaging signals. Existing systems lack a dynamic calibration mechanism based on the response of the entire optical path, resulting in nonlinear fluctuations in the response curve of ultraviolet photon counting under different light intensities. At the same time, optical path fusion calibration often ignores the optical distortion characteristics of the optical system itself, making it impossible to achieve high-precision spatiotemporal synchronization between the spot distribution of ultraviolet imaging and the edge area of ​​visible light images under long focal length and wide field of view conditions. This results in a lack of consistency in detection data when comparing across time periods and instruments.

[0005] 3. Lack of end-to-end real-time adaptive fusion feedback mechanism. Existing optical path calibration processes typically rely on offline calibration or simplified perspective transformation matrices, lacking the ability to adjust in real-time for specific target scenarios. When shooting distance, optical zoom ratio, or external environmental parameters change, the system cannot autonomously sense the optical path offset, nor can it perform real-time self-correction of the image fusion results. This "static calibration, dynamic monitoring" mode results in image fusion accuracy failing to meet sub-pixel-level positioning requirements when facing high-precision tasks such as UHV equipment inspection, thus limiting the ability to deeply analyze discharge trajectory morphology recognition and potential weak insulation hazards.

[0006] To address the aforementioned issues, there is an urgent need for a calibration system based on ultraviolet imaging optical path fusion for corona discharge of power equipment. This system would enable high-precision dynamic registration of multiple sensors in complex environments, real-time compensation for optical distortion, and deep fusion of discharge signals and spatial geometric features, thereby ensuring the accuracy and reliability of locating potential hazards in power equipment. Summary of the Invention

[0007] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a calibration system for ultraviolet imaging optical path fusion based on corona discharge of power equipment, which solves the problems mentioned in the background section.

[0008] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a power equipment corona discharge ultraviolet imaging optical path fusion calibration system, the system comprising an ultraviolet dot matrix emission module, a spatial reference coordinate module, a fusion deviation detection module, an axial error analysis module, a fusion compensation module, and an automatic calibration control module, wherein: The ultraviolet dot matrix emission module includes a housing, a light-transmitting hole, a rectifier module, and LED ultraviolet lamp beads. The three LED ultraviolet lamp beads are arranged in an equilateral triangle topology and are used as an ultraviolet dot matrix signal source to radiate and simulate the spectral characteristics of partial discharge of power equipment. The spatial reference coordinate module is used to generate a theoretical spatial reference coordinate system based on the physical relative positions of the three LED ultraviolet lamp beads; The fusion deviation detection module is used to acquire the fused image of the ultraviolet channel and the visible light channel collected by the multispectral imager under test, and to extract the actual pixel fusion coordinates of each target point. The axial error analysis module is used to perform multi-point topological matching between the actual pixel fusion coordinates and the theoretical spatial reference coordinates, and decouples the registration translation deviation in the X-axis direction and the optical anisotropic distortion parameters in the Y-axis direction through a two-dimensional fusion error separation calibration model. The fusion compensation module is used to calculate the cross-channel spatial mapping reconstruction deviation matrix based on the decoupled registration translation deviation and optical anisotropic distortion parameters. The automatic calibration control module is used to dynamically write the spatial mapping reconstruction deviation matrix into the digital image fusion and stitching engine of the multispectral imager via a wireless data link, and to correct the registration homography matrix in the multi-axis fusion algorithm in real time.

[0009] Preferably, the surface of the housing has a fixing slot, and the surface of the housing is provided with hexagonal head bolts. Three light-transmitting holes are opened on the surface of the housing. The rectifier module is fixedly installed inside the housing by hexagonal head bolts, and the rectifier module is used to provide a constant current to three LED ultraviolet lamp beads. The three LED ultraviolet lamp beads are fixedly installed on the surface of the rectifier module, and the three LED ultraviolet lamp beads are aligned with the three light-transmitting holes one by one. The emission wavelength of the LED ultraviolet lamp beads is 270nm. A Type-C power receiving module is provided on the side of the housing.

[0010] Preferably, in the equilateral triangle topology formed by the three LED ultraviolet lamp beads, the parallel direction of any side is used as the X-axis direction reference, and the direction of the midline pointing from the corresponding vertex to the midpoint of the side is used as the Y-axis direction reference. The axial error analysis module performs independent decoupling calculations on the horizontal stretching ratio and vertical shearing parameter in the fused image through a two-dimensional fusion error separation calibration model.

[0011] Preferably, the front of the housing has visible light contrast crosshairs surrounding the outer side of each light-transmitting hole, and the intersection of the visible light contrast crosshairs coincides with the geometric center of the light-transmitting hole. The rectifier module includes a constant current and voltage regulated drive circuit, which is used to isolate the input voltage fluctuation of the external power supply connected to the Type-C power receiver module and keep the light intensity output fluctuation rate of the three LED ultraviolet lamp beads within a preset threshold of 0.05%.

[0012] Preferably, the fusion deviation detection module has a built-in sub-pixel level edge extraction unit, which is used to perform Gaussian surface fitting on the three ultraviolet spots in the fused image to obtain high-precision sub-pixel level actual pixel fusion centroid coordinates.

[0013] Preferably, the ultraviolet dot matrix emission module controls three LED ultraviolet lamp beads arranged in an equilateral triangle topology to emit stably, and radiates an ultraviolet calibration point source that simulates the partial discharge characteristics of high-voltage power equipment through three light-transmitting holes on the front of the housing. Sp2: The spatial reference coordinate module extracts the absolute center distance of the three LED ultraviolet lamp beads on the physical calibration plane, establishes a two-dimensional standard geometric constraint relationship, and maps it to the theoretical spatial reference coordinate system; Sp3: The fusion deviation detection module guides the multispectral imager under test to align with the physical calibration plane at a preset distance, acquires the ultraviolet spot pixel image captured by the ultraviolet channel and the target surface image captured by the visible light channel, and dynamically fuses and superimposes the two images to extract the actual pixel fusion coordinates of the three target points in the fused image. Sp4: The axial error analysis module performs multi-point topological matching between the actual pixel fusion coordinates and the theoretical spatial reference coordinates in the spatial reference coordinate module, and performs spatial transformation analysis based on the two-dimensional fusion error separation calibration model to solve the registration translation deviation in the X-axis direction and the optical anisotropic distortion parameters in the Y-axis direction. Sp5: The fusion compensation module uses the least squares method to fit and solve the cross-channel spatial mapping reconstruction deviation matrix based on the decoupled registration translation deviation and anisotropic distortion parameters. Sp6: The automatic calibration control module dynamically writes the spatial mapping reconstruction deviation matrix into the digital image fusion and stitching engine of the multispectral imager via a wireless data link, and corrects the registration parameters in the multi-axis fusion algorithm in real time to achieve pixel-level automatic calibration closed loop.

[0014] Preferably, extracting the actual pixel fusion coordinates in Sp3 specifically includes the following sub-steps: Sp3.1: The background light suppression filtering algorithm is used to remove noise from the ultraviolet channel component in the fused image and extract the rough pixel contours of the three ultraviolet spots. Sp3.2: An edge detection operator is used to identify the corresponding visible light contrast crosshair lines in the fused image and to calculate the pixel coordinates of the intersection points of each crosshair line; Sp3.3: Aligns and verifies the center of the rough pixel outline with the corresponding intersection pixel coordinates to obtain the actual pixel fusion coordinates after fusion and overlay.

[0015] Preferably, the invocation of the built-in two-dimensional fusion error separation calibration model in Sp4 specifically includes the following sub-steps: Sp4.1: Convert the theoretical space reference coordinates into homogeneous coordinate form and construct a standard topological matrix; Sp4.2: Align the extracted actual pixel fusion coordinates with the standard topological matrix through an affine transformation to separate the horizontal shearing component and the vertical projection deformation component. Sp4.3: Establish independent error evaluation equations, and calculate the registration translation deviation in the X-axis direction and the optical anisotropic distortion parameters in the Y-axis direction based on the horizontal shear component and the vertical projection deformation component, respectively.

[0016] Preferably, the spatial mapping reconstruction deviation matrix in Sp5 is solved by the inverse perspective transformation function. The decoupled X-axis and Y-axis error terms are used as constraint boundary conditions. The least squares fitting iteration is performed until the reprojection residual is less than the preset pixel threshold, and the final spatial mapping reconstruction deviation matrix is ​​output.

[0017] Preferably, after Sp6, the following closed-loop verification step is also included: Sp7: The automatic calibration control module directs the multispectral imager to re-acquire the corrected fused image. If the recalculated cross-channel pixel centroid deviation is less than the preset zero-five pixels, a calibration success signal is output; otherwise, the reconstruction process is automatically triggered and the process returns to Sp3.

[0018] Beneficial effects This invention provides a light path fusion calibration system for ultraviolet imaging based on corona discharge of power equipment. It has the following advantages: 1. By introducing a spatial geometric calibration mechanism, this invention can effectively eliminate the physical parallax between the ultraviolet imaging detector and the visible light camera, and compensate for the nonlinear geometric distortion caused by changes in equipment attitude. This solution upgrades the traditional fixed parameter mapping to dynamic feature matching, ensuring that the sparse ultraviolet discharge array can be accurately and stably mapped onto the target entity in the visible light image. This effectively solves the problem of misjudgment of hidden dangers caused by spatial mapping deviation, enabling maintenance personnel to accurately locate the core discharge part.

[0019] 2. By constructing a full-link optical path response model, this invention can dynamically calibrate the optical path transmission characteristics in real time according to different lighting conditions, shooting distances, and meteorological characteristics. This mechanism effectively suppresses the interference of strong sunlight and highly reflective backgrounds on ultraviolet imaging, corrects the impact of optical system edge distortion on imaging quality, and achieves high spatiotemporal consistency between ultraviolet discharge spots and visible light images under cross-instrument and cross-environment conditions, greatly improving the reliability and repeatability of corona discharge detection data for power equipment.

[0020] 3. This invention breaks through the limitations of traditional "offline calibration and static monitoring". By introducing a real-time feedback adjustment algorithm, the system has the ability to autonomously sense the optical path status and update the registration strategy in real time when the shooting parameters change. This adaptive fusion mechanism ensures that the image registration maintains sub-pixel-level positioning accuracy throughout the inspection process. It not only meets the high-sensitivity capture of weak insulation hazards in complex environments, but also provides accurate data support for the subsequent in-depth feature analysis and morphological recognition of corona discharge trajectories. Attached Figure Description

[0021] Figure 1 This is a system composition diagram of the present invention; Figure 2 This is a flowchart of the fusion calibration method of the present invention; Figure 3 This is a schematic diagram of the structure of the ultraviolet dot matrix emission module of the present invention; Figure 4 This is a schematic diagram of the rectifier module of the present invention; Figure 5 This is a display of the system main interface of the present invention. Figure 1 ; Figure 6 This is a display of the system main interface of the present invention. Figure 2 .

[0022] 1. Housing; 2. Light-transmitting hole; 3. Rectifier module; 4. LED UV lamp bead; 5. Type-C power receiver module; 6. Mounting slot; 7. Hex head bolt. Detailed Implementation

[0023] 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. Specific Implementation Example 1: Please see Figure 1 and Figure 2 As shown, the ultraviolet imaging optical path fusion calibration system based on corona discharge of power equipment includes an ultraviolet dot array emission module, a spatial reference coordinate module, a fusion deviation detection module, an axial error analysis module, a fusion compensation module, and an automatic calibration control module. Specifically: The ultraviolet (UV) dot matrix emission module is the physical foundation and signal source of the entire calibration system. This module is deployed in the hardware structure of a portable mobile test emission device. In the field of spectral inspection, the ultraviolet signals generated by the corona discharge of high-voltage power equipment are usually weak, transient, and randomly diffused, and cannot be directly used as a quantitative calibration reference. The UV dot matrix emission module, through hardware circuitry and precision mechanical assembly, forces the generation of three UV calibration point sources in three-dimensional physical space with constant spatial relative distance and highly stable energy output. This simulates the core optical characteristics of partial discharge of electrical equipment at a 1:1 scale, providing a high-quality physical model that can be perceived by optical sensors for all subsequent digital registration and reconstruction algorithms.

[0025] Please see Figure 3 and Figure 4As shown, to achieve a battery-free, lightweight design and highly stable optical output, the module is precisely deployed in physical space as follows: The module mainly consists of a housing 1, a light-transmitting aperture 2, a rectifier module 3, and three LED ultraviolet lamp beads 4. It is deployed on a calibration test bench or portable handheld terminal to transmit standard ultraviolet calibration signals to the device under test. The housing 1 serves as the module's carrier, with pre-reserved fixing slots 6 on its surface. The housing 1 not only supports the overall structure but also provides optical shielding and environmental isolation, preventing stray light from entering the module and providing physical protection for the internal electrical components. To prevent the LED UV lamp beads 4 from shifting due to squeezing or dropping, the fixing slot 6 provides a physical positioning reference for the rectifier module, while the hexagonal head bolts 7 ensure that the rectifier module 3 will not experience any slight displacement after being fixed, ensuring that the three LED UV lamp beads 4 and the three light-transmitting holes 2 maintain an absolute geometric alignment relationship in three-dimensional space. The rectifier module 3 is precisely fixed in the slot by the hexagonal head bolts 7 to ensure that the signal transmitter has extremely high mechanical stability during long-term operation. Its core functions are power purification and constant current drive. The rectifier module converts the raw voltage signal with fluctuations and ripples input from the Type-C interface into a signal with extremely low ripple. The stable constant current output, with its internally integrated constant current and voltage regulating drive circuit, effectively blocks external power supply fluctuations, preventing instantaneous voltage drops or surges in external power supply equipment from interfering with the terminal light-emitting element. This fundamentally ensures the lifespan of the light source and the constancy of its luminous power. This circuit can monitor and adjust the current flowing to the LED ultraviolet lamp bead 4 in real time, controlling the light intensity fluctuation within 0.05%. Without this module, the brightness of the lamp bead will fluctuate with changes in power supply voltage, causing the imager to be unable to accurately capture the edge of the point source, thus causing calibration errors. The base of the LED ultraviolet lamp bead 4 is fixedly connected in parallel to the substrate circuit of the rectifier module 3, and is in an equal position. The system features a triangular topological structure, which is the core feature of the system. It provides unique geometric constraints to achieve full-dimensional calibration of the imager's spatial coordinate system (translation, rotation, scaling). As the core signal emission source, it efficiently performs electro-optical energy conversion and emits an ultraviolet light source with a center wavelength strictly limited to 270nm. Its function is to accurately simulate the solar-blind ultraviolet light band unique to the corona discharge of high-voltage power equipment. At the same time, by utilizing its rigid spatial relative spacing, the geometric constraints of the X and Y axes are physically hard-coded into space, thereby realizing the quantitative mapping of the imager's X and Y axis distortion and rotation angle.The light-transmitting aperture 2 is located on the front of the housing 1 and is completely aligned with the LED ultraviolet lamp bead 4 on the optical axis. This design ensures the directional nature of the ultraviolet signal during spatial transmission and reduces stray light interference. This structure serves to collimate the optical path and constrain the space. Since the light emitted by the LED lamp bead has a certain divergence angle, the light-transmitting aperture, through aperture limitation and edge trimming, ensures that the emitted beam exhibits highly directional point source characteristics. This is a key hardware guarantee for ensuring accurate subsequent geometric topology extraction. On the outer surface of the light-transmitting aperture 2, a contrast crosshair is precisely drawn for visual alignment. This involves constructing an internal darkroom environment to isolate stray light and using a spatial aperture to limit the diffused beam emitted by the LED, causing it to converge into a standard point source signal when radiating outwards. The Type-C power receiver module 5 is deployed on the side of the housing 1 at the corresponding opening. Its function is to serve as a standardized physical interface for external universal power supply links, seamlessly receiving external power from smartphones, power banks, or power adapters. This effectively eliminates the device's reliance on bulky lithium batteries, minimizing the physical size and weight of the transmitting device and achieving portability for on-site testing.

[0026] When the power company's maintenance personnel connect the external power supply line to the Type-C power receiver module 5 and close the power control switch, the system is instantly activated from a static state, the hardware current flow channel is opened, and the execution method process Sp1 is formally triggered: Phase 1: Constant Current and Voltage Regulation Drive External power is fed into the rectifier module 3 through the Type-C power receiver module 5. The constant current and voltage regulation drive circuit inside the rectifier module 3 regulates and adjusts the input voltage. Its chip captures the current change in the current circuit in real time through the internal sampling resistor and uses the internal hardware amplifier for negative feedback regulation to force the current output to the three LED ultraviolet lamp beads 4 to be clamped at the preset standard operating current value. Because the output voltage of external power supplies such as smartphones and ordinary power banks often has high-frequency voltage ripple and fluctuation due to the presence of internal chopper circuits, and LEDs are current-sensitive components, even small voltage changes will cause the light emission lumen to flicker violently in a nonlinear high frequency. If it is not rectified and stabilized, when the ultraviolet imager captures the light spot in the subsequent steps, the uneven radiation energy between image frames will cause the sub-pixel centroid extraction algorithm to produce a serious quantization jump phase difference.

[0027] Specifically, let the real-time fluctuation voltage of the input rectifier module be... The precision low-temperature drift sampling resistor integrated inside the rectifier module has a resistance value of The hardware reference voltage source is a stable scalar. The driving circuit outputs a constant driving current to the three parallel / serial LED UV lamp beads 4. The following strongly constrained closed-loop equations must be satisfied: ; After determining a stable drive current Then, the total ultraviolet optical power flux radiated outward by a single LED ultraviolet lamp bead 4 per unit time This can be expressed as the following linear electro-optic energy conversion equation: ; in, The core control objective is to output a constant drive current to the light-emitting diode. The inherent electro-optical conversion efficiency factor (constant at a constant operating temperature); The forward conduction voltage drop of LED UV lamp bead 4 is cut off through hardware feedback. With input fluctuation voltage The algebraic relationship between them results in the final output ultraviolet optical power flux. It exhibits no high-frequency drift over time, achieving absolute photon radiation steady-state in the time domain. Through constant current and voltage regulation, the impact of input voltage fluctuations on the stability of ultraviolet light output can be reduced, thereby improving the spot stability during subsequent image acquisition. The input data is a one-dimensional time-varying voltage signal with voltage glitches and fluctuations from an external interface. The comparator amplifier inside the rectifier chip compares the real-time voltage divider value generated across the sampling resistor with... Continuous high-frequency dynamic comparison is performed. Through hardware-level differential amplification, the on-resistance of the internal power MOSFET is adjusted in real time. The voltage ripple is canceled by the hardware-level "energy absorption-discharge" mechanism, and a DC hard characteristic curve, i.e., a constant current data signal, is output to the subsequent light-emitting load. .

[0028] Phase Two: Stable Ultraviolet Array Emission With rectifier module 3 successfully converting the constant current in the time-domain steady state... The three LED ultraviolet lamp beads 4 are continuously supplied with rated energy and begin to synchronously perform physical light emission. Under the excitation of constant current, the semiconductor energy bands inside the three LED ultraviolet lamp beads 4 undergo electron-hole recombination, which excites stimulated emission and synchronously and stably generates an ultraviolet photon stream with a center wavelength of 270nm. Since its physical base is strictly limited to an equilateral triangle topology during assembly, these three photon streams have an inherent spatial relative geometric distance from the beginning of their generation. The "270nm wavelength" was chosen because this wavelength is precisely within the ultraviolet light band generated by partial discharge of high-voltage electrical equipment and belongs to the "solar blind zone" of the Earth's atmosphere, which can effectively eliminate the interference of stray sunlight in nature and ensure that calibration can be carried out under sunlight. The "equilateral triangle topology" design is to provide both the X-axis spatial constraint in the reference side direction and the Y-axis spatial constraint in the corresponding midline direction when establishing the mathematical benchmark in the next step. This provides a stable physical topological basis for subsequently isolating and calibrating the two-dimensional axial deformation error of the multispectral imager.

[0029] Phase 3: Spatial Limiting Output of Ultraviolet Point Source: Because bare LED chips typically have a wide viewing angle (e.g., a 120° Lambert beam), if spatial confinement and ignition are performed through the light-passing aperture 2, the emitted ultraviolet light will undergo extensive overlap and scattering in space, even causing secondary reflection interference on the shell surface. The physical constraint of the light-passing aperture 2 allows the diffused light clusters to be confined and condensed into isolated point source signals with extremely clear spatial geometric contours and well-defined boundaries. This ensures that subsequent multispectral or imaging equipment can obtain images with extremely high contrast when capturing this signal, providing the sub-pixel centroid algorithm with extreme precision for analysis. Therefore, the three LEDs arranged in an equilateral triangle emit light outwards... After emitting highly stable 270nm ultraviolet photons, the diffused light beam travels forward and encounters the physical obstruction of the inner wall of the light-transmitting hole 2 on the front of the housing 1, and begins spatial beam shaping. The diffused light beam is directed outward through the three light-transmitting holes 2 on the front of the housing 1. The inner wall of the light-transmitting hole 2 physically cuts off the large-angle diverging light at the edge (beam clipping). Finally, only the collimated ultraviolet light with extremely high straightness that propagates vertically forward along the center line of each lamp's optical axis passes through the housing, forming three pure ultraviolet calibration point sources in the three-dimensional space directly in front of the device, which perfectly coincide with the center of the crosshairs on the outer side.

[0030] With the execution of Sp1, the ultraviolet dot matrix emission module successfully deployed three 270nm ultraviolet calibration point source entities in physical space that are highly stable, have pure spectral bands, and possess standard equilateral triangle topological geometric constraints. However, these standard physical topological dot matrices are still only physical objects at this point. In order for the system to have a computable digital standard answer in subsequent computer vision comparisons, the system needs to translate these physical center distance mechanical dimensions into matrices in the digital world. Therefore, the system will automatically generate a control flow state transition instruction, smoothly transition to the next module, call the spatial reference coordinate module to execute Sp2, and formally start the mapping and establishment process of the theoretical spatial reference coordinate system.

[0031] The spatial reference coordinate module is the mathematical definition engine and digital standard measurement of this calibration system. This module is deployed on the calibration instrument's main control computer, central processing unit (CPU), or embedded system-on-chip (SoC). The main function of this module is to establish an absolute mathematical and geometric reference and eliminate spatial nonlinear ambiguity. When the multispectral imager is tested on site, due to hand shaking or tilting of the field of view, the acquired images will rotate, translate, and scale at arbitrary angles. If directly compared, the system cannot distinguish which deformations are caused by defects in the lens itself and which are caused by the shooting angle. The spatial reference coordinate module establishes an absolutely standard, immutable two-dimensional theoretical coordinate system in the digital world by reading the physically hard-coded mechanical dimensions, so that subsequent error analysis has an absolutely objective alignment template.

[0032] This module employs the following precise deployment in its logic registers and physical storage to ensure strict mapping between hardware and software data: Hard-coded physical parameter registers: Deployed in the non-volatile flash memory (Flash) or EEPROM of the main control chip, this structure stores the absolute physical center distance parameters (i.e., the side lengths of the equilateral triangles) of the ultraviolet dot matrix emission module during precision mechanical assembly at the factory. Its function is to serve as the source of the entire mathematical derivation, eliminating errors caused by human measurement; Coordinate matrix generation processor: Deployed in the floating-point arithmetic unit (FPU) inside the processor, its function is to execute mathematical instructions such as square root, matrix construction and homogeneous coordinate transformation at high speed, ensuring the establishment of spatial reference in milliseconds.

[0033] In order to transform the physical topology in Sp1 into a mathematical template that the algorithm can directly call, the system automatically generates a control flow jump instruction to enter Sp2: Phase 1: Extracting the Physical Absolute Center Distance: After the spatial reference coordinate module is started, the coordinate matrix generator first sends a data retrieval instruction to the physical parameter hard-coded register to directly extract the absolute center distance data of the three LED UV beads 4 pre-embedded in it on the physical calibration plane. Based on the hard-coded design of the transmitter, the side length of the equilateral triangle is defined as... .

[0034] The second stage: Establishing two-dimensional standard geometric constraints: In a two-dimensional plane, any three non-collinear points possess unique geometric determinism. Using the parallel direction of any side as the X-axis reference and the median direction from the corresponding vertex to the midpoint of that side as the Y-axis reference, this maximizes the utilization of the high symmetry and stable geometric topology of equilateral triangles. This geometric constraint eliminates the translational and rotational degrees of freedom of the three-point array in space, ensuring that each spatial coordinate point inside the triangle possesses a unique and standardized mathematical analytical expression. Therefore, in obtaining... After obtaining the numerical values, the processor dynamically constructs a standard Cartesian two-dimensional rectangular coordinate system in the digital space. The constraint construction rules are strictly limited as follows: the parallel direction of any selected side of the equilateral triangle formed by the three LED ultraviolet lamp beads is used as the X-axis direction reference, and the definition of the reference side remains unchanged throughout the calibration process. The direction of the median pointing from the corresponding vertex to the geometric midpoint of the side is used as the Y-axis direction reference, and the geometric midpoint of the selected reference side is established as the origin of the coordinate system. .

[0035] Phase 3: Mapping the Theoretical Spatial Reference Coordinate Matrix: To transform the aforementioned two-dimensional geometric constraints into a computer matrix flow, the system invokes the built-in topological rigidity calibration model. This model utilizes analytical geometry principles to calculate the theoretical absolute coordinate matrix of the three vertices in the homogeneous coordinate system. Specifically, firstly, based on the geometric relationship of an equilateral triangle, the two-dimensional spatial reference height in the Y-axis direction is calculated. : ; Based on the origin defined above The model accurately derives the first vertex of the triangle. Second vertex Third vertex Two-dimensional theoretical space coordinates: , , The model combines these three discrete coordinate points into one. Theoretical space reference coordinate matrix : ; With the theoretical space reference coordinate matrix The successful construction and output in Sp2 established a perfect answer in the digital world. However, this standard answer currently only exists in memory. In order to quantify the actual imaging deviation of the multispectral imager under test in the real world, the system needs to instruct the imager to capture physical signals. Therefore, the system automatically flows and generates the control signal for the next step, enters Sp3, and calls the fusion deviation detection module to guide the multispectral imager under test to perform dynamic fusion acquisition of two heterogeneous channel images and extraction of actual pixel coordinates.

[0036] The fusion deviation detection module is the image data acquisition center and pixel-level feature extraction engine of this calibration system. This module is deployed as a logic algorithm unit in the main control computer, central processing unit (CPU) or dedicated image signal processor (ISP) of the calibrator. It is connected to the data output end of the multispectral imager under test in real time through a data communication interface (such as a high-speed digital bus or wireless video transmission link). Since the multispectral imager contains two completely independent hardware optical path systems, the ultraviolet channel and the visible light channel, the output image format, resolution and spatial coordinates are completely heterogeneous when facing the same physical calibration plane. The fusion deviation detection module is to pull the two asynchronously acquired optical image streams to the same time and space dimension for dynamic superposition. Through advanced computer vision algorithms, it converts the blurred and diffuse physical light spots in the image into high-precision pixel coordinate data, eliminating the physical barrier that cross-channel images cannot be directly compared. This provides the most original and accurate pixel-level measured coordinate samples for subsequent modules to analyze optical axis deviation.

[0037] To achieve efficient cross-channel image processing and pixel feature extraction, this module is deployed in the hardware operating environment and storage space as follows: Image acquisition buffer register: Deployed in the system's high-speed dynamic random access memory (DRAM), this structure consists of an ultraviolet image buffer and a visible light image buffer. Its function is to synchronously latch the high frame rate, uncompressed heterogeneous raw image data transmitted from the imager, and prevent frame misalignment caused by hardware delay in the multi-channel data stream. Digital image fusion and stitching engine: Deployed in the processor's graphics computing unit (GPU) or dedicated digital signal processor (DSP), its function is to run pixel-level dynamic registration algorithm, perform real-time weight addition and matrix remapping of two heterogeneous fields of view, and thus synthesize a digital fused image; Subpixel-level edge extraction unit: Deployed as a software structure in the core execution layer of the algorithm, its role is specifically to perform subpixel-level geometric center localization of target points in the fused image.

[0038] Since a theoretical benchmark independent of external deformation was generated in Sp2, the system guides the hardware peripherals to acquire the latest physical spot image in order to compare the difference between the physical image and the theoretical model. Therefore, the system automatically triggers and enters Sp3: Phase 1: Guiding the multispectral imager under test to align with the physical calibration plane at a preset distance: After the fusion deviation detection module is activated, it first outputs a set of dynamic calibration auxiliary frames on the system's user interface or the imager's display screen through the built-in guidance protocol. The system guides the operator or automated robotic arm to place the multispectral imager under test at a preset distance (preferably 3-5 meters) perpendicular to the physical calibration plane of the ultraviolet dot matrix emission module. At this time, the point source signals emitted by the three LED ultraviolet lamp beads 4 on the calibration plane fall exactly into the center of the field of view of the multispectral imager. The preset distance of "3-5 meters" and "vertical alignment" are to ensure that the physical calibration plane occupies the best pixel area ratio on the imager's sensor (CMOS / CCD) and ensure that both optical channels are within their optimal depth-of-field focusing range, thereby greatly reducing the peripheral linear perspective tearing caused by large-angle tilt and ensuring that the acquired original pixels have the highest signal-to-noise ratio.

[0039] Phase 2: Acquire the ultraviolet spot pixel image captured by the ultraviolet channel and the target surface image captured by the visible light channel: After the imager completes the alignment and locking of the physical field of view, the optical characteristics of ultraviolet light and visible light are completely different. In order to preserve the purest original optical characteristics of the two channels, the system immediately triggers the dual-channel synchronous exposure mechanism. The automatic calibration control module directs the dual sensors of the multispectral imager to work in parallel. The ultraviolet channel inside captures the 270nm wavelength ultraviolet light spot emitted by three LED ultraviolet lamp beads through a specific filter and CMOS sensor, generating a high-contrast ultraviolet channel pixel image. Simultaneously, the visible light channel is also activated, independently capturing the target background image on the calibration plane (including the physical outline of the calibration plane and the contrast crosshairs), generating a visible light channel pixel image. These two independent image data streams are then fed into the dynamic image acquisition buffer register.

[0040] Third stage: Dynamically fuse and overlay the two images: Once the two independent channel images are secured in the image acquisition buffer register, in order to represent the relative misalignment of the optical axes in the same unified coordinate space, the system immediately sends the data stream into the digital image fusion and stitching engine. The engine calls the dual-channel image spatial weight superposition and registration model of the inner cover, using the visible light image... As the background image for the entire space, use ultraviolet images. As a feature foreground layer, the spatial weighted overlap registration model performs spatial-level pixel fusion of images from heterogeneous channels through linear weight mapping. Specifically, let the fused digital image be... The mathematical fusion equation is as follows: ; in, and The preset image channel fusion weight coefficients satisfy... (Preferably, set) (to highlight the characteristics of ultraviolet light spots). The initial spatial registration homogeneous transformation matrix is ​​used to perform the initial scaling and alignment. The input is two two-dimensional image matrices synchronously latched in a register. and The graphics processing unit (GPU) first processes the ultraviolet image through a parallel rendering pipeline. Initial spatial matching of the matrix, followed by matching the same coordinates of the two images. The pixel grayscale values ​​at each location are multiplied and added using the formula above to output a full-color, dynamically fused image that precisely overlays the invisible ultraviolet spot onto the visible light target background. .

[0041] Fourth stage: Extract the actual pixel fusion coordinates of the three target points in the fused image: The generation of dynamically fused images ensures that heterogeneous light spots and background target points exist in the same pixel coordinate system. To quantize these visual features into coordinate vectors that can be mathematically solved, the system automatically activates sub-pixel-level edge extraction units. Sp3.1: Background Light Suppression Filtering and Coarse Ultraviolet Spot Contour Extraction: In actual test environments, there are often stray light noises such as residual sunlight and wall reflections. If background light is not suppressed, the edges of the ultraviolet spot will adhere to these background noises and become fuzzy and deformed, resulting in severe nonlinear coarsening in subsequent center calculations. Therefore, the system first extracts the fused image. The ultraviolet channel component is processed by calling the built-in adaptive morphological background suppression algorithm to filter out low-frequency non-uniform background light and high-frequency impulse noise. After highlighting the spot area, a local adaptive threshold operator is used to transform the image into a binary matrix, and connected component tracing is performed along the edge pixels to extract the rough pixel contours of the three ultraviolet spots. Specifically, morphological top-hat transformation is first used to suppress the ultraviolet component. The low-frequency background grayscale is used to obtain a background-suppressed image. : ; In the formula, This indicates a morphological opening operation (erosion followed by dilation). The model uses pre-defined regular structural elements based on the expected pixel size of the light spot. Then, it calls the adaptive Otsu method to dynamically calculate the local adaptive segmentation threshold. Binarization cropping was performed to extract the rough pixel contour boundary point set vectors of the three ultraviolet spots. (in The data input is the two-dimensional ultraviolet pixel grayscale matrix decomposed from the fused image. GPUs utilize structuring elements Parallel opening operations are performed to estimate the background trend, and high-pass filtering and binarized topological boundary tracking are achieved through matrix subtraction. The output is a set of coarse pixel contour boundary points for three isolated ultraviolet spots.

[0042] SP3.2: Visible Light Crosshair Recognition and Intersection Calculation: Since physical crosshairs are drawn on the outer side of the light-transmitting hole 2 on the front of the emitter housing 1, and the physical geometric center of the light-transmitting hole 2 is rigidly locked to coincide with the optical axis of the LED ultraviolet lamp bead 4 during design and manufacturing, the geometric intersection of the drawn crosshairs is completely coaxial with the physical center of the ultraviolet emission source in three-dimensional space. This allows the visible light channel to capture this crosshair intersection, essentially equivalent to capturing the absolute projection point of the ultraviolet point source on the visible light sensor, forming the core mechanical geometric center reference in the visible light channel. Only by accurately finding the pixel coordinates of the intersection points of these three sets of crosshairs in the image coordinate system using computer vision algorithms can... It can obtain the actual projection point of the visible light optical axis on the sensor, thus providing an absolute visible light spatial anchor point for subsequent cross-channel coaxiality alignment verification. Therefore, after SP3.1 successfully locked the rough pixel outlines of the three ultraviolet spots and completed the digital discretization of ultraviolet features, the module activated its visual edge detection operator to perform high-precision line edge retrieval on the visible light channel components in the fused image, identify the visible light contrast crosshair lines drawn on the outer side of the convex front light-transmitting hole, and accurately calculate the pixel coordinates of the intersection points of these three sets of scale lines through a set of analytical geometric equations. Specifically, after edge detection, the horizontal scale lines fitted in each light-transmitting hole region using the Hough linear transform are... and vertical scale lines The analytical equations are as follows: ; The analytical model obtains the pixel coordinates of their intersection point in the fused image coordinate system by simultaneously solving the above equations. : ; The input is the two-dimensional visible light channel background component matrix of the fused image. The algorithm first performs Gaussian smoothing and gradient magnitude calculation to extract step edge points. Then, it transforms the edge points to polar coordinate parameter space and uses an accumulator to vote and fit a straight line equation. Finally, it solves for the intersection points through algebraic simultaneous equations. The output is three sets of pixel coordinates of the intersection points of the visible light contrast crosshairs corresponding to the three light apertures. .

[0043] Sp3.3: Centroid-Intersection Alignment Verification and Actual Pixel Fusion Coordinate Reconstruction: By performing a corresponding alignment verification (i.e., coaxial residual comparison) between the geometric center of the ultraviolet spot and the cross intersection of the visible light, abrupt gross errors caused by single-channel instantaneous image tearing, jitter, or edge noise interference can be effectively eliminated. Therefore, the sub-pixel-level feature extraction unit extracts the geometric centroid of each rough pixel contour in Sp3.1, and performs corresponding point alignment and rigidity consistency verification with the pixel coordinates of the corresponding cross scale line intersection calculated in Sp3.2. After passing the confidence level verification, the verified and reconstructed coordinates are used as the actual pixel fusion coordinate matrix after final fusion and superposition. Output, for the first output of Sp3.1 The rough outline of the ultraviolet spot area The model first calculates the coordinates of its geometric contour center (centroid) using first-order pixel image moments. : ; Subsequently, the verification algorithm introduces a strong Euclidean distance constraint equation to compare the contour centers. Intersection Does the spatial residual meet the set rigidity consistency threshold? : ; After the residual passes the threshold check, the system sets the final actual pixel fusion coordinates. The points are bound to this high-confidence location and arranged into the corresponding sequence of the measured matrix, which is the actual pixel fusion coordinate matrix of the three target points under the current imaging system. : ; With high-precision, high-confidence actual pixel fusion coordinate matrix With the successful reconstruction and final output in SP3.3, this calibration system has simultaneously grasped the theoretical standard answer at the data flow level. ) and physical measured performance ( In order to analyze the specific physical defect components that cause these misalignments (such as optical axis translation and lens anisotropic distortion), the system enters Sp4 and calls the axial error analysis module to solve the multi-point topology matching and two-dimensional fusion error separation calibration model.

[0044] The axial error analysis module is the geometric decoupling brain and physical error separation core of this calibration system. This module is deployed as a high-priority algorithm component in the arithmetic logic layer of the calibration system's central processing unit (CPU) or embedded system-on-chip (SoC). In actual multispectral imager manufacturing and inspection applications, optical axis translation parallax caused by non-parallel machining of dual channels and anisotropic optical distortion caused by defects in aspherical lens processing are often intertwined and superimposed in the acquired two-dimensional images. If they are not decomposed, the subsequent compensation algorithm will fall into the local extremum dilemma of multidimensional nonlinear optimization, resulting in a calibration dead loop. The axial error analysis module performs spatial transformation analysis based on the two-dimensional fusion error separation calibration model. Using multivariate affine transformation registration and orthogonal projection technology, it scientifically decomposes the intertwined mixed spatial deformation into registration translation deviation in the X-axis direction and optical anisotropic distortion parameters in the Y-axis direction. This cuts off the coupling interference between errors in different dimensions, thereby providing completely decoupled and physically meaningful underlying parameter support for the subsequent fine reconstruction matrix.

[0045] To ensure the stable and efficient execution of high-precision decoupling operations, this module employs the following dedicated hardware / software mapping structures at the logic and register levels: Topology matching register cluster: Deployed in the system's high-speed temporary storage area, its function is to latch and align the point-to-point mapping relationship between actual pixel coordinates and theoretical reference coordinates in parallel, and to play a role in preventing the reversal of point position numbers in spatial multi-point topology matching. Affine registration and orthogonal projection operation unit: As a high-priority microcode operator cluster, it is embedded in the processor's arithmetic logic pipeline. Its role is to perform low-level matrix algebra operations such as affine transformation alignment, matrix least squares orthogonal projection, and matrix inversion, ensuring that the decoupling speed reaches the millisecond level. Axial error independent component buffer: Deployed in on-chip shared memory, its function is to safely store the independent deviation parameters of the decoupled output, and it acts as a bridge to stably transmit the registration translation deviation and optical anisotropic distortion parameters to the subsequent fusion compensation module.

[0046] In Sp3, the system transforms the measured point matrix of the physical world into a discrete pixel coordinate matrix. In order to compare these measured coordinates one by one with the answers established in Sp2, the system automatically triggers and enters Sp4: Phase 1: Homogeneous Dimensional Upgrading of Theoretical Coordinates and Construction of Standard Topological Matrices The axial error analysis module first uses a homogeneous planar space dimensionality-upgrading mapping model to upgrade the two-dimensional theoretical plane reference coordinates from the previous step by introducing homogeneous constant terms. The coordinates are then matrixed and stacked column by column according to the standard topological geometric order to finally construct a theoretical standard topological matrix. Since translation is a vector addition and rotation and scaling are matrix multiplications in ordinary planar coordinate systems, complex errors cannot be uniformly represented. After being converted into a homogeneous matrix, all errors such as misalignment, shearing, and distortion can be completely unified into a single matrix multiplication form, achieving efficient parallel separation of multiple parameters.

[0047] The planar homogeneous space updimensional mapping model introduces a homogeneous coordinate system, transforming the nonlinear affine transformation in the two-dimensional plane into linear matrix operations in a high-dimensional space, thereby locking the rigid topology. The theoretical planar coordinates of the three ideal target points output by Sp2 are as follows: , and The dimension-upgrading mapping model transforms planar coordinates into homogeneous coordinates by appending a scalar 1 to the end of each coordinate vector. Then, the homogeneous vectors of the three target points are sequentially stacked into a matrix to reconstruct the theoretical standard topological matrix. ;in, The horizontal and vertical coordinates of the ideal target point A in the digital reference space; : The horizontal and vertical coordinates of the ideal target point B in the digital reference space; : Horizontal and vertical coordinates of the ideal target point C in the digital reference space; 1: Homogeneous scaling constant factor used for translation transformation, where the input data is a two-dimensional theoretical space reference coordinate matrix carrying the ideal geometric spacing. (for one) (The algebraic matrix), the data computation unit initiates the dimension-upgrading operator, appends a hardware complement factor to the end of each coordinate pair, and performs matrix transpose and concatenated stacking to output a homogeneous theoretical standard topological matrix. (for one) (a homogeneous square matrix).

[0048] Phase Two: Alignment of Multivariate Affine Matrix and Decoupling of Geometric Deformation Components Because slight axial torsion and non-uniformity errors in lens processing often exist during the assembly of dual-channel imaging devices, these mechanical and optical errors are intertwined. Optimal approximation through affine transformation is necessary to mathematically isolate all deformation components. The module then directly calls the actual pixel fusion coordinate matrix extracted by Sp3. After transposing it, it is compared with the standard topology matrix constructed in the first stage. Affine transformation alignment is performed, and the optimal affine mapping operator is obtained through least-squares fitting. From this, the horizontal shear component and the vertical projection deformation component, representing the physical phase structure and surface deformation, are accurately separated. Specifically, the transpose of the actual pixel fusion coordinate matrix is ​​expressed as follows: (composed of the pixel coordinates of three actual target points stacked in columns) The spatial affine mapping relationship between two sets of topological matrices can be expressed as: ; in, For the solution to be found The eigenparametric structure of the α-order synthetic affine transformation matrix is ​​decomposed as follows: ; in, : Horizontal axis scaling factor, which represents the asymmetric scaling factor of the image in the horizontal direction; Vertical axis scaling factor, representing the asymmetric scaling factor of the image in the vertical direction; The separated horizontal shearing component characterizes the degree of non-orthogonal distortion experienced by the cross-channel image in the horizontal direction; : The separated vertical projection deformation component, which characterizes the physical deformation degree of the image along the longitudinal optical axis due to perspective deformation and nonlinear stretching; : These represent the global spatial translation of the image in the horizontal and vertical directions (unit: pixels). The algorithm constructs a residual sum of squares to minimize the objective function, and then directly obtains the optimal closed-form solution of the affine operator using generalized inverse matrix multiplication. ; Using this analytical solution, the module can instantly extract the horizontal shear component, which contains translational and torsional features. With vertical projection deformation component It can be completely and independently separated and extracted from complex composite measured coordinate errors.

[0049] Phase 3: Establishment of Independent Error Equations and Anisotropy Classification of Biaxial Correction Values The physical adjustment mechanisms and underlying firmware of multispectral imagers are typically controlled independently along the hardware axis. Without axial decoupling, calibration will cause coordinate shifts in the other axis during single-axis adjustment, leading to oscillations during the calibration process. Therefore, the module establishes independent error evaluation equations and calculates the registration translation deviation in the X-axis direction and the optical anisotropic distortion parameters in the Y-axis direction based on the horizontal shear component and vertical projection deformation component, respectively. This set of equations is precisely categorized and summarized according to the physical error generation mechanism: the spatial translation term representing the global lateral displacement caused by the parallax between the ultraviolet and visible light optical axes is normalized and combined, and refined... The registration translation deviation is converted into the X-axis direction. Simultaneously, the characteristic parameters representing multiaxial non-uniform stretching and perspective projection variations caused by anisotropic distortion of lens materials or surface shapes are polynomial-fitted and refined into optical anisotropic distortion parameters in the Y-axis direction. This ultimately forms two completely decoupled and non-interfering anisotropic axial calibration data streams. Specifically, by establishing an algebraic mapping equation between the error source and the physical adjustment axis, the coupled affine deformation matrix elements are completely decoupled into one-dimensional discrete calibration scalars. Based on the affine matrix obtained in the second stage... The model is constructed using the following decoupled equation system. First, for the spatial translation term of the global displacement caused by the parallax of the two optical axes, the horizontal shear component separated in the second stage is introduced into the algebraic equation for normalization and mapping compensation, thereby directly solving for the registration translation deviation in the X-axis direction: ; Secondly, to address the nonlinear deformation caused by lens surface processing errors, the vertical projection deformation component separated in the second stage and the horizontal and vertical scaling difference term are introduced into the coupled constraint equation for nonlinear deformation fitting, thereby directly calculating the optical anisotropic distortion parameters in the Y-axis direction: ; in, The global horizontal translation amount obtained in the second stage; The horizontal and vertical scaling factors obtained in the second stage; The horizontal shear component separated in the second stage; The vertical projection deformation component separated in the second stage; : represents the preset calibration gain mapping coefficient. This set of equations successfully achieved physical isolation of error sources and axial independent solution.

[0050] With the axial error analysis module successfully establishing and inversely solving the independent error evaluation equation in Sp4, it robustly outputs the registration translation deviation in the X-axis direction and the optical anisotropic distortion parameters in the Y-axis direction. The entire calibration system has thoroughly understood the optical and physical defects of the multispectral imager at the algorithm level. In order to transform these independent error parameters of each axis into an image reconstruction network that can correct the dynamic video stream in real time, the system control flow automatically flows and smoothly enters Sp5, calling the fusion compensation module. Based on the decoupled registration translation deviation and optical anisotropic distortion parameters, the final cross-channel spatial mapping reconstruction deviation matrix is ​​solved by least squares fitting.

[0051] The fusion compensation module serves as the execution bridge for the entire calibration system, moving from mathematical analysis to physical engineering applications. As the core data processing and reconstruction engine, it is logically deployed within the central processing unit of the calibration system or the embedded graphics computing board of the multispectral imager. The registration translation deviation in the X-axis direction and the optical anisotropic distortion parameters in the Y-axis direction output by the axial error analysis module are discrete physical characteristic parameters that cannot be directly applied to real-time dynamic video streams containing millions of pixels. The role of the fusion compensation module is to fuse these discrete single-axis deviation parameters into a spatial mapping reconstruction deviation matrix that can perform real-time coordinate transformation on all image pixels through an advanced mathematical model. This solves the problems of image edge tearing and center misalignment caused by non-parallel optical axes and lens anisotropic distortion in heterogeneous channels (ultraviolet and visible light), achieving pixel-level full alignment.

[0052] In the physical deployment of hardware and data, this module is connected to the front-end axial error analysis module through an internal high-speed data bus to dynamically receive decoupled error data. Its output is connected to the digital image fusion and stitching engine inside the multispectral imager through a data communication interface (such as a wireless data link or a dedicated system bus) to form a closed-loop link. The module consists of a boundary condition constraint unit, a matrix iteration calculation unit, and a reprojection residual evaluation unit. Each unit works together through microcode instructions to ensure that the matrix solution process is completed within milliseconds.

[0053] In Sp4, the system divides the aliased mounting parallax and lens distortion into registration translation deviation in the X-axis direction and optical anisotropic distortion parameters in the Y-axis direction. This provides excellent initialization boundaries for high-precision matrix construction. In order to transform these independent axial parameters into cross-channel reconstruction instructions for the entire image, the system enters Sp5, and the specific implementation process is as follows: Phase 1: Constructing the inverse perspective transformation function model: Because the two channels (ultraviolet and visible light) of a multispectral imager are physically located differently, they are essentially two different perspective projection planes when imaging the same physical calibration plane. To achieve pixel-level perfect overlap of the two images, a one-to-one mapping relationship between the two planes needs to be established. Ordinary linear addition cannot eliminate the nonlinear warping at the lens edge. Therefore, an inverse perspective transformation function is used to construct a reconstruction matrix (i.e., homography matrix) with eight degrees of freedom, which can simultaneously correct translation, rotation, scaling, shearing, and perspective distortion.

[0054] Define the cross-channel spatial mapping reconstruction deviation matrix as follows: Its mathematical expression is as follows: ; in, : Represents the spatial mapping reconstruction deviation matrix across channels, which is manifested as an anti-distortion fusion correction matrix at the physical hardware execution level; : These represent the horizontal scaling mapping element and the horizontal shearing mapping element in the first row of the matrix, respectively; : Represents the horizontal translation mapping element of the first row of the matrix; : These represent the vertical shearing mapping element and the vertical scaling mapping element in the second row of the matrix, respectively; : Represents the vertical translation mapping element in the second row of the matrix; : Represents the horizontal perspective projection deformation constraint element and the vertical perspective projection deformation constraint element in the third row of the matrix, respectively; 1: Represents the constant normalization factor in the lower right corner of the matrix, the theoretical space reference coordinate point. Coordinates fused with actual pixels The following homogeneous space mapping relationship is established using the inverse perspective transformation function: Expanding this into a system of inverse perspective transformation nonlinear mapping equations in a nonhomogeneous planar coordinate system, it can be expressed as: ; ; in, : Represents the index of the calibration target point, which can be selected as in this scheme. One target point; : respectively represent the first The horizontal and vertical theoretical space reference coordinates of an ideal target point in the digital reference space; : These represent the first and second parts extracted from the fused image by the previous step. The actual pixel horizontal fusion coordinates and the actual pixel vertical fusion coordinates of each target point; : These represent the corresponding algebraic transformation characteristic terms in the spatial mapping reconstruction deviation matrix.

[0055] Second stage: Introducing boundary constraints and least squares fitting iteration: In actual measurements, the sensor itself contains thermal noise, and image noise can cause slight jitter (pixel burrs) in the actual pixel coordinates of the extracted target points. If the equations are solved directly using algebraic methods, the noise will be amplified, resulting in severe distortion in image areas other than the calibration points. To improve the robustness of the calibration system, this invention uses the registration translation deviation in the decoupled X-axis direction and the optical anisotropic distortion parameters in the Y-axis direction as constraint boundary conditions to limit the reconstruction matrix parameters. The search range is defined, and the least squares method is introduced for global iterative fitting. The goal is to find a set of optimal matrix coefficients that minimizes the sum of squared cumulative errors of all calibrated target points after reprojection. The core optimization formula for least squares fitting is: ; Meanwhile, the optimization objective equation is strongly constrained by the decoupling boundary conditions of the preceding axial error: , ,in, : indicates the use of spatial mapping to reconstruct the deviation matrix The optimal solution for each element is the objective, which minimizes the subsequent objective algebraic expression. : Indicates the ideal target Target With target Perform summation and accumulation operations on the projected residuals; : indicates calculating the square of the L2 norm of the multidimensional vector within the parentheses, which is the square of the Euclidean distance between the measured coordinates and the projected reconstructed coordinates; : indicates the first Each target point is a two-dimensional measured vector composed of the horizontal and vertical components of the actual pixel fused coordinates; : indicates the first Each target point is a two-dimensional theoretical vector composed of the horizontal and vertical components of the theoretical space reference coordinates; : Represents the inverse perspective transformation projection function model performed on the theoretical coordinate vector based on the spatial mapping reconstruction deviation matrix; : Represents a preset horizontal axis error extraction function, used to extract the components in the matrix element that are related to horizontal spatial misalignment; : Represents a preset vertical nonlinear error extraction function, used to extract the anisotropic components in the matrix element that are related to the stretching of the lens surface; X-axis registration translation deviation: represents the horizontal rigid displacement dissociated by the axial error analysis module in Sp4; Optical anisotropic distortion parameter in the Y-axis direction: represents the longitudinal nonlinear perspective stretching deformation dissociated from the axial error analysis module in Sp4.

[0056] The input stream consists of two parts. The first part is the actual pixel fusion coordinates of the three target points extracted by Sp3 from the fused image. Theoretical spatial reference coordinates generated by Sp2 The second part consists of the registration translation error in the X-axis direction and the optical anisotropic distortion parameters in the Y-axis direction, output by the Sp4 decoupling inverse solution. The computation unit first substitutes the input data into the aforementioned constrained least squares optimization equation. The system iterates using the Gauss-Newton iteration method or the Levenberg-Marquardt (LM) nonlinear optimal solution operator. In each iteration, the reprojection residual evaluation unit calculates the current reprojection residual (i.e., the Euclidean distance between the reconstructed point and the actual point). The fitting iteration process continues until the reprojection residual is less than a preset pixel threshold (e.g., 0.1 pixels), at which point the final optimal solution is output. After convergence, the module outputs the final cross-channel spatial mapping reconstruction deviation matrix with anti-distortion and anti-parallax functions. That is, the final anti-distortion fusion correction matrix.

[0057] The output of the spatial mapping reconstruction deviation matrix signifies that the system has completed the thorough quantification and mathematical reconstruction of the physical defects of the optical path at the algorithm level. However, this matrix is ​​currently stored in the calibration system's memory as a set of optimized mathematical coefficients. In order for these coefficients to truly play a role and correct the optical image of the imager under test in real time, the system must send them down and write them into the imager's hardware chip. Since the fusion compensation module has already output the final anti-distortion fusion correction matrix through least squares fitting iteration, in order to apply the mathematical reconstruction result to the actual optical device, the system will automatically generate a connection command and transfer it to Sp6, calling the automatic calibration control module to perform dynamic matrix writing and real-time calibration closed loop through the wireless data link.

[0058] The automatic calibration control module is the highest command and state machine coordination unit of this calibration system. This module adopts a distributed deployment architecture across hardware platforms. Its main logic is deployed on the main control chip of the calibration instrument, while its communication peripherals and execution terminals are seamlessly deployed on the embedded microprocessor and wireless communication baseboard of the multispectral imager. After the front-end modules complete the arduous image acquisition, geometric decoupling, and matrix fitting, all optimization results are stored in the calibration instrument's temporary register. The role of the automatic calibration control module is to break down the boundaries between hardware and software, dynamically writing the spatial mapping reconstruction deviation matrix into the digital image fusion and stitching engine of the multispectral imager via a wireless data link. This allows for real-time correction of the registration homography matrix in the multi-axis fusion algorithm, achieving fully automatic pixel-level multi-axis registration. Simultaneously, it also undertakes full-process quality monitoring. Its responsibility is to ensure that the final image fusion accuracy strictly meets the standard through a closed-loop verification algorithm, and to force a process rollback when the standard is not met, so as to achieve truly unattended closed-loop calibration. In terms of physical hardware deployment, the module includes a wireless data link chip installed inside the housing 1 and connected to the main control chip, as well as a matching radio frequency antenna. The wireless data link chip usually adopts a high-bandwidth, low-latency Wi-Fi or Bluetooth communication module. At the multispectral imager end, the module is directly connected to the configuration register bus of the digital image fusion and stitching engine (usually composed of a high-parallelism FPGA chip or a dedicated DSP) inside the imager. The module is composed of a wireless protocol serialization unit, a register direct mapping driver, and a state machine state controller. The units are interconnected to ensure highly reliable flow of control commands and matrix data.

[0059] Since Sp5 calculates the perfect cross-channel spatial mapping reconstruction bias matrix, it provides a complete mathematical foundation for real-time hardware correction of the multispectral imager. To enable the static matrix parameters to produce a real-time correction effect on the dynamic video stream, the system enters Sp6: In industrial-grade power line inspection, multispectral imagers need to output fused video of ultraviolet and visible light at a rate of no less than 30 frames per second. If the calibration matrix remains only at the application software layer, the large system latency in processing image data by general-purpose CPUs will cause severe ghosting and misalignment in the fused image. Therefore, the spatial mapping reconstruction deviation matrix needs to be directly written into the underlying hardware fusion engine of the imager. Dynamic writing via wireless data link can eliminate cumbersome physical wiring and enable maintenance personnel to achieve remote, non-contact, fully automatic, and real-time calibration at the substation site. After receiving the matrix, the digital image fusion and stitching engine calls the registration homography matrix hardware acceleration execution model. This model uses the input spatial mapping reconstruction deviation matrix (i.e., the registration homography matrix) to perform pixel-by-pixel projective transformation remapping on the input ultraviolet image in the FPGA pipeline architecture.

[0060] Specifically, let the input raw UV channel pixel coordinates be a vector. The transformed pixel coordinates are The elements of the spatial mapping reconstruction deviation matrix (registration homography matrix) are derived from... The mathematical model for hardware remapping is as follows: ; Expanding to a system of algebraic equations that can be directly executed by the hardware pipeline, we get: ; ; in, : These represent the original horizontal and vertical pixel coordinates of any discrete pixel in the original high-frequency image stream acquired and output in real time by the ultraviolet image sensor in a multispectral imager. : These represent the horizontal and vertical corrected pixel coordinates of the corresponding pixels in the new ultraviolet channel video stream, which is completely coaxially registered with the visible light channel optical axis after real-time repositioning and correction by the digital image fusion and stitching engine. : These represent the eight optimal constant matrix element parameters of the spatial mapping reconstruction deviation matrix that are written into the configuration register in reverse via the wireless data link.

[0061] The input data stream consists of two parts. The first part consists of nine parameters of the spatial mapping reconstruction deviation matrix, which are serialized and directly written into the configuration register by the automatic calibration control module via a wireless data link. Secondly, the original high-frequency image pixel stream acquired and output in real time by the ultraviolet image sensor is used to read the coordinates of one pixel in each clock cycle of the digital image fusion and stitching engine. The engine directly calls the hardware multiplier and pipeline divider inside the chip and substitutes them into the above-mentioned hardware-accelerated execution model formula for coordinate repositioning. Since the repositioned coordinates are usually non-integers, the system synchronously uses a bilinear interpolation algorithm to perform fusion and synchronous calculation on the gray value of the target pixel to eliminate the jagged edges of the image. The system outputs a new video stream of the ultraviolet channel that has been corrected in pixel spatial position and is fully coaxially registered with the optical axis of the visible light channel. This completes the real-time dynamic correction of the registration homography matrix and registration parameters in the digital image fusion and stitching engine of the multispectral imager.

[0062] When Sp6 successfully injects the optimal spatial mapping reconstruction deviation matrix into the high-speed configuration register of the imager, and the hardware video fusion stitching engine completes the real-time dynamic correction of the registration homography matrix in the multi-axis fusion algorithm, the cross-channel fused image output by the imager has theoretically achieved pixel-level spatial alignment. However, in order to ensure that the fused image output by the system truly meets the quantitative calibration requirements of corona discharge of power equipment under complex interference conditions such as at substation sites, the system must perform fully automatic quality inspection on the real image after updating parameters. To this end, the control state machine pointer is extended backward, forcibly triggering the Sp7 closed-loop verification step.

[0063] The core system module for performing this step is the automatic calibration control module, which is specifically executed by its internally integrated closed-loop verification unit. This module is deployed using a distributed cross-platform architecture and is responsible for quantitative final quality testing of the actual fused image after hardware parameter updates. This solves the problem of partial failure of calibration parameters caused by slight fluctuations in environmental wind speed, thermal deformation of equipment, or occasional packet loss in wireless transmission in traditional open-loop calibration systems. It ensures that the multispectral imager finally delivered to power operation and maintenance personnel has absolutely reliable and traceable pixel-level fusion accuracy.

[0064] To achieve the aforementioned closed-loop control, the module is precisely configured with the following auxiliary hardware and semiconductor mapping structures at the physical and logic layers: Wireless communication bus interface: Deployed in the peripheral circuit of the calibration instrument microprocessor, it is specifically used to send retake and reacquisition commands to the multispectral imager in reverse, and to capture the latest fusion feature data stream fed back by the imager in real time; State machine control register: A non-volatile control storage area deployed inside the main control chip, used to store the current hardware control status code (such as "waiting for retake", "residual iteration", "output success", "force rollback"). Its function is to strictly control the flow and timing of the calibration process through hardware-level instruction jumps. Non-volatile parameter solid-state memory (flash memory / EEPROM): Deployed on the internal circuit board of the multispectral imager, its function is to enable hardware write protection only when a "calibration successful" signal is received from the automatic calibration control module, and permanently solidify the current spatial mapping reconstruction deviation matrix to prevent parameter loss after the device is powered off.

[0065] In actual calibration operations, external factors such as wind speed fluctuations, support tremors, or minor measurement errors in calibration distance can all cause residual deviations in the matrix parameters calculated in a single operation. Without final effect testing, the system will not be able to determine whether the currently output fused image has truly reached the deliverable accuracy standard. The automatic calibration control module directs the multispectral imager to re-acquire the corrected fused image and feeds the calibration results back to the deviation extraction front end for secondary quantitative evaluation. This can eliminate occasional environmental interference and achieve pixel-level automatic calibration closed loop for the entire system.

[0066] After hardware correction and update of the image, the cross-channel pixel centroid deviation evaluation model is used to re-quantitatively assess whether the cross-channel fusion error has converged. The system re-extracts the actual pixel coordinates of the three target points in the corrected fused image. Let the re-extracted and updated ultraviolet spot sub-pixel centroid coordinates be a vector. The corresponding pixel coordinates of the intersection of the visible light contrast crosshairs are Its cross-channel pixel centroid deviation The comprehensive evaluation formula for Euclidean distance is as follows: ; in, : This represents the cross-channel pixel centroid deviation (unit: pixels) recalculated by the system after hardware parameter updates, used to measure the current actual quality of the fused image. : This represents the arithmetic mean of the pixel-level spatial reprojection residuals of the three validation target points; : Indicates the target index for re-validation, respectively : These represent the re-extraction of the 1st, 2nd, and 3rd elements from the updated fused image, as directed by the automatic calibration control module. The horizontal and vertical centroid coordinates of the center of each ultraviolet spot sub-pixel; : These represent the corresponding in the updated fused image. The pixel coordinates of the horizontal and vertical intersection points of the visible light contrast crosshair; : This represents the two-dimensional geometric Euclidean distance between the recalibrated ultraviolet centroid and the visible light intersection point under the same target.

[0067] The input consists of the corrected fused image re-acquired and output by the multispectral imager after the hardware parameters have been updated, as directed by the automatic calibration control module, and three sets of new channel centroid pixel coordinate pairs calculated a second time by the fusion deviation detection module. The state machine controller calls the aforementioned evaluation and judgment model to recalculate the cross-channel pixel centroid deviation. After the calculation is completed, the control logic executes the conditional branch threshold determination: the current threshold is set... The value is compared with a strict threshold preset by the system, which is strictly limited to 0.5 pixels.

[0068] Data Output (Branch 1: Calibration Successful): If the recalculated cross-channel pixel center deviation is less than the preset 0.5 pixels (i.e. If the cross-channel misalignment is less than 0.5 pixels, a calibration success signal will be output: In the field of discharge detection of high-voltage power equipment, cross-channel misalignment of less than 0.5 pixels is completely imperceptible to the eye. The accuracy of the overlap between the discharge spot and the outer contour of the equipment has reached an extremely high level. At this time, the optical axis alignment has completely eliminated lens distortion and mechanical installation parallax. Further optimization will cause the algorithm to fall into overfitting. Specifically, the comparator outputs a high-level trigger signal, and the automatic calibration control module then sends a solidification configuration command to the multispectral imager. After receiving the command, the imager firmware immediately opens the write protection and injects the spatial mapping reconstruction deviation matrix in the current digital image fusion and splicing engine into its internal non-volatile parameter solidification memory. At the same time, the green LED indicator on the surface of the calibration instrument lights up, and the user interface outputs the prompt "Calibration successful, current cross-channel fusion error is less than 0.5 pixels". The state machine control register is set to "output successful" and is safely suspended, indicating that the calibration closed loop has been successfully completed.

[0069] Data Output (Branch 2: Automatically Trigger Reconstruction Process and Backtrack): If the recalculated cross-channel pixel center deviation is greater than or equal to the preset 0.5 pixels (i.e. If the deviation is greater than or equal to 0.5 pixels, it means that during the matrix calculation or wireless transmission process, it was affected by occasional physical noise such as strong electromagnetic pulses from substations, sudden strong winds causing vibrations of the testing platform, or strong ultraviolet light interference from the background environment. This caused the current correction matrix to fail to completely eliminate cross-channel deformation. If the calibration is ended directly, it will lead to false reports of the actual spatial location of partial discharge during subsequent inspections, causing potential power safety hazards. Therefore, it is necessary to start over and implement adaptive environmental noise filtering by resampling. Specifically, the comparator outputs a low-level trigger signal to automatically calibrate and control the process. The module's state machine controller immediately initiates the abnormal interrupt rollback mechanism. The module first sends a clear command to the system register to completely discard the spatial mapping reconstruction deviation matrix calculated in the current round, preventing dirty data from lingering. Then, the control module sends a synchronous reset signal to the multispectral imager via the wireless data link, commanding it to clear the abnormal register value in the stitching engine. The state machine control register forcibly jumps to "forced rollback", automatically triggering the reconstruction process and returning to Sp3. The multispectral imager is then commanded to re-acquire new dynamic fused images and extract the actual pixel fusion coordinates, starting a new round of iterative optimization until the residual perfectly converges to within 0.5 pixels.

[0070] By implementing fully automatic dynamic parameter correction and quality closed-loop management for Sp6 and Sp7 through the automatic calibration control module, this invention has successfully built a complete fully automatic closed-loop control ecosystem in the field of power equipment power detection, from physical emission, digital decoupling, matrix fitting to hardware injection and residual verification. It has effectively overcome the industry technical bottleneck of difficulty in real-time elimination of cross-channel alignment hysteresis and nonlinear deformation of multiple video streams.

[0071] For further details, please refer to Figure 5 and Figure 6 As shown, it provides a system main interface display diagram, which includes a system status display area, a calibration control selection area, and a calibration result display area. It is used to realize real-time registration monitoring of ultraviolet and visible light images of corona discharge of power equipment, selection and execution of automated calibration strategies, and intuitive verification and closed-loop feedback of fusion deviation compensation effect.

[0072] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0073] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A calibration system for ultraviolet imaging optical path fusion based on corona discharge of power equipment, characterized in that: The system includes an ultraviolet dot matrix emission module, a spatial reference coordinate module, a fusion deviation detection module, an axial error analysis module, a fusion compensation module, and an automatic calibration control module, wherein: The ultraviolet dot matrix emission module includes a housing (1), a light-transmitting hole (2), a rectifier module (3), and LED ultraviolet lamp beads (4). The three LED ultraviolet lamp beads (4) are arranged in an equilateral triangle topology and are used as an ultraviolet dot matrix signal source to radiate and simulate the spectral characteristics of partial discharge of power equipment. The spatial reference coordinate module is used to generate a theoretical spatial reference coordinate system based on the physical relative positions of the three LED ultraviolet lamp beads (4); The fusion deviation detection module is used to acquire the fused image of the ultraviolet channel and the visible light channel collected by the multispectral imager under test, and to extract the actual pixel fusion coordinates of each target point. The axial error analysis module is used to perform multi-point topological matching between the actual pixel fusion coordinates and the theoretical spatial reference coordinates, and decouples the registration translation deviation in the X-axis direction and the optical anisotropic distortion parameters in the Y-axis direction through a two-dimensional fusion error separation calibration model. The fusion compensation module is used to calculate the cross-channel spatial mapping reconstruction deviation matrix based on the decoupled registration translation deviation and optical anisotropic distortion parameters. The automatic calibration control module is used to dynamically write the spatial mapping reconstruction deviation matrix into the digital image fusion and stitching engine of the multispectral imager via a wireless data link, and to correct the registration homography matrix in the multi-axis fusion algorithm in real time.

2. The optical path fusion calibration system for ultraviolet imaging based on corona discharge of power equipment according to claim 1, characterized in that: The surface of the housing (1) is provided with a fixed slot (6), the surface of the housing (1) is provided with a hexagonal head bolt (7), the three light-transmitting holes (2) are opened on the surface of the housing (1), the rectifier module (3) is fixedly installed inside the housing (1) by the hexagonal head bolt (7), and the rectifier module (3) is used to provide a constant current for the three LED ultraviolet lamp beads (4). The three LED ultraviolet lamp beads (4) are fixedly installed on the surface of the rectifier module (3), and the three LED ultraviolet lamp beads (4) are aligned with the three light-transmitting holes (2). The emission wavelength of the LED ultraviolet lamp beads (4) is 270nm. The side of the housing (1) is provided with a Type-C power receiving module (5).

3. The optical path fusion calibration system for ultraviolet imaging based on corona discharge of power equipment according to claim 2, characterized in that: In the equilateral triangle topology formed by the three LED ultraviolet lamp beads (4), the parallel direction of any side is taken as the X-axis direction reference, and the direction of the center line pointing from the corresponding vertex to the midpoint of the side is taken as the Y-axis direction reference. The axial error analysis module performs independent decoupling calculations on the horizontal stretching ratio and vertical shearing parameter in the fused image through the two-dimensional fusion error separation calibration model.

4. The optical path fusion calibration system for ultraviolet imaging based on corona discharge of power equipment according to claim 2, characterized in that: The front of the housing (1) is provided with visible light contrast cross scale lines surrounding the outer side of each light-transmitting hole (2). The intersection of the visible light contrast cross scale lines coincides with the geometric center of the light-transmitting hole (2). The rectifier module (3) includes a constant current voltage regulation drive circuit, which is used to isolate the input voltage fluctuation of the external power supply connected to the Type-C power receiving module (5) and keep the light intensity output fluctuation rate of the three LED ultraviolet lamp beads (4) within a preset threshold of five ten-thousandths.

5. The optical path fusion calibration system for ultraviolet imaging based on corona discharge of power equipment according to claim 1, characterized in that: The fusion deviation detection module has a built-in sub-pixel level edge extraction unit, which is used to perform Gaussian surface fitting on the three ultraviolet spots in the fused image to obtain high-precision sub-pixel level actual pixel fusion centroid coordinates.

6. The optical path fusion calibration system for ultraviolet imaging based on corona discharge of power equipment according to claim 1, characterized in that: The fusion calibration method of the system includes the following steps: Sp1: The ultraviolet dot matrix emission module controls the three LED ultraviolet lamp beads (4) arranged in an equilateral triangle topology to emit stably and radiate outward through the three light-transmitting holes (2) on the front of the housing (1) to simulate the ultraviolet calibration point source of the partial discharge characteristics of high-voltage power equipment. Sp2: The spatial reference coordinate module extracts the absolute center distance of the three LED ultraviolet lamp beads (4) on the physical calibration plane, establishes a two-dimensional standard geometric constraint relationship, and maps it to the theoretical spatial reference coordinate system; Sp3: The fusion deviation detection module guides the multispectral imager under test to align with the physical calibration plane at a preset distance, acquires the ultraviolet spot pixel image captured by the ultraviolet channel and the target surface image captured by the visible light channel, and dynamically fuses and superimposes the two images to extract the actual pixel fusion coordinates of the three target points in the fused image. Sp4: The axial error analysis module performs multi-point topological matching between the actual pixel fusion coordinates and the theoretical spatial reference coordinates in the spatial reference coordinate module, and performs spatial transformation analysis based on the two-dimensional fusion error separation calibration model to solve the registration translation deviation in the X-axis direction and the optical anisotropic distortion parameters in the Y-axis direction. Sp5: The fusion compensation module uses the least squares method to fit and solve the cross-channel spatial mapping reconstruction deviation matrix based on the decoupled registration translation deviation and anisotropic distortion parameters. Sp6: The automatic calibration control module dynamically writes the spatial mapping reconstruction deviation matrix into the digital image fusion and stitching engine of the multispectral imager via a wireless data link, and corrects the registration parameters in the multi-axis fusion algorithm in real time to achieve pixel-level automatic calibration closed loop.

7. The optical path fusion calibration system for ultraviolet imaging based on corona discharge of power equipment according to claim 6, characterized in that: Extracting the actual pixel fusion coordinates from Sp3 specifically includes the following sub-steps: Sp3.1: The background light suppression filtering algorithm is used to remove noise from the ultraviolet channel component in the fused image and extract the rough pixel contours of the three ultraviolet spots. Sp3.2: An edge detection operator is used to identify the corresponding visible light contrast crosshair lines in the fused image and to calculate the pixel coordinates of the intersection points of each crosshair line; Sp3.3: Aligns and verifies the center of the rough pixel outline with the corresponding intersection pixel coordinates to obtain the actual pixel fusion coordinates after fusion and overlay.

8. The optical path fusion calibration system for ultraviolet imaging based on corona discharge of power equipment according to claim 6, characterized in that: The specific steps involved in calling the built-in two-dimensional fusion error separation calibration model in Sp4 are as follows: Sp4.1: Convert the theoretical space reference coordinates into homogeneous coordinate form and construct a standard topological matrix; Sp4.2: Align the extracted actual pixel fusion coordinates with the standard topological matrix through an affine transformation to separate the horizontal shearing component and the vertical projection deformation component. Sp4.3: Establish independent error evaluation equations, and calculate the registration translation deviation in the X-axis direction and the optical anisotropic distortion parameters in the Y-axis direction based on the horizontal shear component and the vertical projection deformation component, respectively.

9. The optical path fusion calibration system for ultraviolet imaging based on corona discharge of power equipment according to claim 6, characterized in that: The solution of the spatial mapping reconstruction deviation matrix in Sp5 is achieved by the inverse perspective transformation function. The decoupled X-axis and Y-axis error terms are used as constraint boundary conditions. The least squares fitting iteration is performed until the reprojection residual is less than the preset pixel threshold, and the final spatial mapping reconstruction deviation matrix is ​​output.

10. The optical path fusion calibration system for ultraviolet imaging based on corona discharge of power equipment according to claim 6, characterized in that: The following closed-loop verification steps are included after Sp6: Sp7: The automatic calibration control module directs the multispectral imager to reacquire the corrected fused image. If the recalculated cross-channel pixel centroid deviation is less than the preset zero-five pixels, a calibration success signal is output; otherwise, the reconstruction process is automatically triggered and the process returns to Sp3.

Images