Method and system for detecting micro-cracks of a solvent-free polyester coating anticorrosive layer

Abstract

The present application provides a solvent-free polyester coating anticorrosive layer micro-crack image detection method and system, comprising: sequentially applying heat excitation to the anticorrosive layer to be detected along multiple preset heat excitation directions, collecting and registering infrared time sequence images, and generating a sequence of directional heat response images; extracting the thermal phase value of the pixel neighborhood, constructing a thermal phase gradient vector field, identifying and connecting local saddle points under different preset heat excitation directions, forming a saddle point migration trajectory and a saddle point migration feature set; determining a suspected interface micro-crack area according to the saddle point migration feature set, performing bidirectional thermal scanning including the heating path and the cooling path on the suspected interface micro-crack area; constructing a thermal hysteresis loop based on the bidirectional thermal scanning infrared image sequence, extracting the loop gap feature, the phase transition feature and the path asymmetry feature; and spatially fusing the saddle point migration feature set and the thermal hysteresis feature set to determine a closed interface micro-crack candidate area.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method and system for detecting microcracks in solvent-free polyester coating anticorrosion layers. Background Technology

[0002] Solvent-free polyester coatings are widely used for corrosion protection of metal pipelines, pressure vessels, and steel structural components. During curing, installation, and service, microcracks can easily form inside the coating or at the interface between the coating and the metal substrate due to coating shrinkage, thermal expansion and contraction of the substrate, local residual stress, and temperature cycling. Some microcracks are initially closed and covered by a complete coating, making them difficult to detect visually or through visible light imaging. Once corrosive media enter the coating, they may propagate along these closed interfacial microcracks, causing coating blistering, peeling, and substrate corrosion.

[0003] Existing infrared thermal imaging detection methods typically identify defect areas based on temperature changes, temperature difference distribution, or heat diffusion rate under a single thermal excitation condition, offering the advantage of non-contact detection. However, uneven coating thickness, localized curing differences, interface bubbles, differences in weld thermal conductivity, and changes in substrate curvature can all generate thermal responses similar to those of interface microcracks, easily leading to misjudgments. Especially for closed interface microcracks, the crack walls may remain in contact at room temperature, resulting in weak abnormal responses from a single thermal excitation, making accurate differentiation from non-crack defects difficult.

[0004] Therefore, this invention proposes a method and system for detecting microcracks in solvent-free polyester coating anticorrosive layers. The information disclosed in the background section is only for enhancing understanding of the background of this disclosure and may therefore contain prior art information that is not common knowledge to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a method and system for detecting microcracks in solvent-free polyester coating anticorrosive layers, thereby solving the technical problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for detecting microcracks in solvent-free polyester coating anticorrosive layers includes the following steps: S1. Divide the anti-corrosion layer to be tested into detection units, apply thermal excitation sequentially along multiple preset thermal excitation directions, collect and register the corresponding infrared time sequence images, and generate a sequence of directional thermal response images. S2. Based on the azimuth thermal response image sequence, extract the thermal phase value of the pixel neighborhood, construct the thermal phase gradient vector field, identify local saddle points, and connect the corresponding local saddle points according to the preset thermal excitation azimuth order to form the saddle point migration trajectory and saddle point migration feature set. S3. Select effective saddle point migration trajectories based on the saddle point migration feature set, determine the suspected interface microcrack region around the effective saddle point migration trajectory, and perform bidirectional thermal scanning including heating path and cooling path on the suspected interface microcrack region to acquire bidirectional thermal scanning infrared image sequence. S4. Based on the bidirectional thermal scanning infrared image sequence, extract the local thermal phase response, local thermal amplitude response and local gradient diffusion response, construct a thermal hysteresis loop, extract the loop gap feature, phase transition feature and path asymmetry feature to form a thermal hysteresis feature set. S5. Spatially fuse the saddle point migration feature set and the thermal hysteresis feature set to determine the candidate region of closed interface microcracks, extract the center line of closed interface microcracks, and output the location, extension direction and risk level of closed interface microcracks.

[0007] S1 specifically includes: selecting the area to be inspected on the surface of the anti-corrosion layer to be inspected, dividing the area to be inspected into multiple inspection units, setting multiple thermal excitation units, infrared imaging components and edge computing nodes, and establishing inspection configuration data between inspection units, preset thermal excitation orientation, stable structure position and pixel coordinates; based on the inspection configuration data, sequentially controlling the corresponding thermal excitation units to apply thermal excitation within a safe range according to the preset thermal excitation orientation, and simultaneously acquiring infrared images to form an original orientation thermal response image sequence; performing quality screening and spatial registration on the original orientation thermal response image sequence to generate an orientation thermal response image sequence, and saving the pixel coordinate range of the inspection units.

[0008] S2 specifically includes: reading the azimuth thermal response image sequence, extracting the temperature response curve of the pixel neighborhood for the preset thermal excitation azimuth, and calculating the thermal phase value based on the response delay of the temperature response curve relative to the thermal excitation reference sequence to form a thermal phase map set; filtering the thermal phase map set, calculating the thermal phase gradient vector, constructing the thermal phase gradient vector field, identifying local saddle points based on the horizontal and vertical second-order thermal phase changes, and forming local saddle point records; matching and connecting local saddle points in adjacent azimuths according to the preset thermal excitation azimuth order to form saddle point migration trajectories, and extracting the migration distance, migration direction, trajectory continuity, azimuth response stability, and trajectory coverage pixel coordinate range to form a saddle point migration feature set.

[0009] S3 specifically includes: reading the saddle point migration trajectory and saddle point migration feature set, filtering valid saddle point migration trajectories, expanding and merging overlapping areas around the valid saddle point migration trajectories to form a record of suspected interface microcrack areas; selecting the corresponding thermal excitation unit based on the suspected interface microcrack area record, sequentially increasing and decreasing the heat input per unit area according to the thermal excitation level to form heating and cooling paths; acquiring a baseline local image sequence before the start of bidirectional thermal scanning, and acquiring, filtering, registering, and cropping local infrared images during bidirectional thermal scanning, pairing the heating path image and cooling path image corresponding to the same thermal excitation level and relative sampling time to form a bidirectional thermal scanning infrared image sequence.

[0010] S4 specifically includes: reading the baseline local image sequence and the bidirectional thermal scanning infrared image sequence before scanning; extracting the local thermal phase response, local thermal amplitude response, and local gradient diffusion response according to the thermal scanning path, thermal excitation level, and pixel neighborhood to form a bidirectional local thermal response pair; normalizing and weighting the bidirectional local thermal response pair to obtain the comprehensive thermal response; connecting the comprehensive thermal responses corresponding to the heating path and cooling path according to the heat input per unit area to construct a thermal hysteresis loop; extracting loop gap features, phase transition features, and path asymmetry features based on the thermal hysteresis loop; and statistically analyzing the continuous abnormal pixel neighborhood along the migration trajectory of the main effective saddle point to form a continuous abnormal segment and thermal hysteresis feature set.

[0011] S5 specifically includes: reading the saddle point migration feature set and thermal hysteresis feature set, establishing a correspondence according to the suspected interface microcrack area number, calculating the spatial overlap ratio between the trajectory band area and the continuous abnormal section, generating and merging closed interface microcrack candidate areas; performing connectivity and centerline refinement processing on the closed interface microcrack candidate areas, extracting the closed interface microcrack centerline, converting the actual position, determining the extension direction, and determining the risk level based on the saddle point migration feature set, thermal hysteresis feature set, and spatial overlap ratio; superimposing the closed interface microcrack centerline, actual position, extension direction, and risk level onto the reference infrared image to generate a closed interface microcrack image recognition report.

[0012] The beneficial effects of this invention are as follows: This invention applies thermal excitation sequentially along multiple preset thermal excitation orientations and constructs a thermal phase gradient vector field based on infrared time-series images. It can utilize the local saddle point migration characteristics formed when heat flows around interfacial microcracks to locate hidden microcracks within the anti-corrosion layer or at the interface between the anti-corrosion layer and the metal substrate. This overcomes the limitations of visual inspection and visible light image detection in detecting closed interfacial microcracks. Suspected interfacial microcrack areas are screened based on the saddle point migration trajectory, and bidirectional thermal scanning is performed only on suspected areas, reducing redundant analysis of non-abnormal areas, decreasing the amount of image processing data, and improving the efficiency of large-area anti-corrosion layer detection.

[0013] This invention constructs a thermal hysteresis loop by comparing the local thermal phase response, local thermal amplitude response, and local gradient diffusion response under heating and cooling paths. It then extracts loop notch features, phase transition features, and path asymmetry features to identify changes in crack wall contact state, improving the detection capability of closed-type interface microcracks. Spatially fusing the saddle point migration feature set with the thermal hysteresis feature set, and utilizing the spatial overlap between the trajectory band region and continuous abnormal segments for dual verification, reduces the possibility of misjudging film thickness variations, local curing differences, interface bubbles, and weld thermal conductivity differences as interface microcracks, thus improving the accuracy of detection results.

[0014] This invention extracts the centerline of closed-type interface microcracks and converts the image pixel coordinates into the actual coordinates of the surface of the anti-corrosion layer to be inspected. This allows for the output of the location and extension direction of the closed-type interface microcracks, providing a clear basis for subsequent verification and local repair. By combining saddle point migration distance, trajectory continuity, azimuth response stability, thermal hysteresis characteristics, and continuous distribution length, the risk level is determined, and a closed-type interface microcrack image recognition report is generated. This report can distinguish between undetected states and abnormal detection states, providing a basis for arranging the maintenance sequence of the anti-corrosion layer. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the microcrack image detection method for the anti-corrosion layer according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the intelligent sensing system structure according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the saddle point migration trajectory and suspected region determination in an embodiment of the present invention; Figure 4 This is a schematic diagram of bidirectional thermal scanning fusion recognition according to an embodiment of the present invention. Detailed Implementation

[0016] 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.

[0017] Example 1: As Figure 1 As shown in the figure, this embodiment provides a method for detecting microcracks in a solvent-free polyester coating anti-corrosion layer, including the following steps: S1. Divide the anti-corrosion layer to be tested into detection units, apply thermal excitation sequentially along multiple preset thermal excitation directions, collect and register the corresponding infrared time sequence images, and generate a sequence of directional thermal response images. S2. Based on the azimuth thermal response image sequence, extract the thermal phase value of the pixel neighborhood, construct the thermal phase gradient vector field, identify local saddle points, and connect the corresponding local saddle points according to the preset thermal excitation azimuth order to form the saddle point migration trajectory and saddle point migration feature set. S3. Select effective saddle point migration trajectories based on the saddle point migration feature set, determine the suspected interface microcrack region around the effective saddle point migration trajectory, and perform bidirectional thermal scanning including heating path and cooling path on the suspected interface microcrack region to acquire bidirectional thermal scanning infrared image sequence. S4. Based on the bidirectional thermal scanning infrared image sequence, extract the local thermal phase response, local thermal amplitude response and local gradient diffusion response, construct a thermal hysteresis loop, extract the loop gap feature, phase transition feature and path asymmetry feature to form a thermal hysteresis feature set. S5. Spatially fuse the saddle point migration feature set and the thermal hysteresis feature set to determine the candidate region of closed interface microcracks, extract the center line of closed interface microcracks, and output the location, extension direction and risk level of closed interface microcracks.

[0018] S1 specifically includes the following sub-steps: S110. Select the area to be tested on the surface of the solvent-free polyester coating anti-corrosion layer, divide the area to be tested into multiple adjacent testing units, and set up multiple thermal excitation units (e.g., using an infrared halogen lamp tube with adjustable power or a laser heater), one infrared imaging component (e.g., using an uncooled focal plane infrared thermal imager) and one edge computing node (e.g., using an industrial control computer equipped with a GPU) around the area to be tested. A detection unit is the smallest spatial region used to perform local thermal response analysis. Each detection unit has a unique detection unit number and corresponds to the actual spatial extent of the surface of the anti-corrosion layer to be tested and the range of continuous pixel coordinates in the infrared image. Adjacent detection units share boundaries along the lateral or longitudinal direction to facilitate subsequent calculation of thermal phase changes between adjacent locations.

[0019] The size of the detection unit is determined based on the image resolution of the infrared imaging component, the actual size of the area to be detected, and the preset minimum pixel coverage, ensuring that each detection unit covers at least 16px × 16px in the infrared image. If the detection unit size is too small, the local thermal response is easily affected by the temperature measurement noise of a single pixel; if the detection unit size is too large, the local thermal diffusion anomalies caused by microcracks at the interface are easily masked by the regional average value.

[0020] For example, when using an infrared imaging component with an image resolution of 640px×640px for a detection area with an actual size of 200mm×200mm, the detection area is divided into 20 rows and 20 columns, totaling 400 detection units. The actual size of each detection unit is 10mm×10mm, corresponding to 32px×32px.

[0021] Using the center of the area to be tested as a reference point, the direction in which each thermal excitation unit points towards the center of the area to be tested is defined as the preset thermal excitation orientation. Each thermal excitation unit has a unique thermal excitation unit number, and each preset thermal excitation orientation has a unique orientation number. In this embodiment, eight thermal excitation units are set, with an angular interval of 45° between adjacent preset thermal excitation orientations. The eight preset thermal excitation orientations are 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°, respectively. The thermal excitation unit number, the preset thermal excitation orientation, and the effective thermal excitation coverage of each thermal excitation unit are derived from the installation configuration file generated after the equipment is installed.

[0022] The infrared imaging component acquires a reference infrared image of the area to be detected before thermal excitation begins. Edge computing nodes use image recognition algorithms to extract stable structure locations from the reference infrared image. Stable structure locations are image feature locations that can be repeatedly identified before and after thermal excitation, and whose identification position does not shift due to the thermal excitation process in the absence of device displacement.

[0023] For stable structural positioning, heat-resistant positioning marks set around the periphery of the area to be inspected are preferred. When heat-resistant positioning marks are not set, texture inflection points with fixed structural edges or unchanged positions before and after thermal excitation are used, and texture features inside suspected defect areas are not used. For example, one heat-resistant positioning mark with a diameter of 3mm is set around the perimeter of the area to be inspected, and the pixel coordinates of the centers of the four heat-resistant positioning marks are recorded.

[0024] During the equipment installation and calibration phase, a pixel coordinate to actual coordinate transformation matrix is ​​generated for each detection unit based on its actual spatial range and pixel coordinate range. The edge computing node writes the following information into the detection configuration data: the area to be detected number, detection unit number, actual spatial range of the detection unit, pixel coordinate range of the detection unit, pixel coordinate to actual coordinate transformation matrix, thermal excitation unit number, preset thermal excitation orientation, effective thermal excitation coverage, stable structure position pixel coordinates, image resolution of the infrared imaging component, fixed sampling interval, unified timestamp, detection configuration data version, and calibration parameter version.

[0025] A unified timestamp is generated by the edge computing node and sent to the thermal excitation unit and the infrared imaging component to associate the thermal excitation process with the infrared image acquisition process.

[0026] S120, the edge computing node controls the activation of each thermal excitation unit sequentially according to the location number order based on the detection configuration data. For the d-th thermal excitation unit, its heat input per unit area is calculated using the following formula:

[0027] in, This represents the heat input per unit area corresponding to the d-th thermal excitation unit, in J / cm². This represents the output power of the d-th thermal excitation unit, in W, and is derived from the feedback value of the thermal excitation unit controller. This represents the duration for which the d-th thermal excitation unit continuously applies thermal excitation, in seconds, and is derived from the difference between the start and stop timestamps recorded by the edge computing node. The effective thermal excitation coverage area of ​​the d-th thermal excitation unit is expressed in cm², derived from the equipment installation and calibration results; d represents the thermal excitation unit number.

[0028] The heat input per unit area satisfies:

[0029] in, This represents the minimum heat input per unit area required to generate a recognizable thermal response. This indicates the maximum heat input per unit area that will not cause softening, color change, decreased adhesion, or new cracks in the anti-corrosion layer being tested. and The sample was pre-calibrated using a template that has the same coating system, application thickness, and curing conditions as the anti-corrosion layer to be tested.

[0030] For example, when the output power is 20W, the duration is 5s, and the effective thermal excitation coverage area is 100cm², the heat input per unit area is 1J / cm²; if the safe range is 0.6J / cm² to 1.4J / cm², then the thermal excitation meets the acquisition requirements.

[0031] For each preset thermal excitation orientation, the infrared imaging component sequentially performs baseline acquisition, loading acquisition, and diffusion acquisition. With a fixed sampling interval of 0.1s, 10 frames are acquired before the start of thermal excitation, 50 frames are acquired during the duration of thermal excitation, and 40 frames are acquired during the heat diffusion period after the end of thermal excitation, forming a total of 100 infrared images. Each infrared image records the thermal excitation unit number, preset thermal excitation orientation, image frame number, image frame acquisition time, current thermal excitation state, and current output power. After completing the acquisition for one preset thermal excitation orientation, the edge computing node continuously acquires infrared images of the area to be detected.

[0032] When the difference between the current average temperature of the area to be detected and the average temperature of the area calculated based on the reference infrared image is no higher than 0.3℃, the next thermal excitation unit corresponding to the preset thermal excitation orientation is activated. The infrared images corresponding to each preset thermal excitation orientation are arranged according to the orientation number and the image frame acquisition time to generate the original orientation thermal response image sequence.

[0033] S130, the edge computing node performs quality screening on each frame of the infrared image in the original orientation thermal response image sequence. For the nth frame of the infrared image, the sharpness index and local overexposure ratio are calculated according to the following formula:

[0034] in, This represents the sharpness index of the nth frame of the infrared image; This represents the nth frame of the infrared image; This represents the edge response image obtained by performing a Laplacian operation on the nth frame of the infrared image; Indicates variance operation; This represents the local overexposure ratio of the nth frame of the infrared image; This indicates the number of pixels that reach the upper limit of the infrared imaging component's range; This indicates the total number of pixels in the infrared image frame; n represents the infrared image frame number.

[0035] The sharpness threshold and the local overexposure ratio threshold are derived from the template calibration results. Edge computing nodes simultaneously compare the acquisition time difference between two adjacent infrared images. When the acquisition time difference deviates from the fixed sampling interval by more than 0.02 seconds, the corresponding time period is marked as an abnormal sampling period.

[0036] After quality screening, affine registration is performed on each frame of infrared images using stable structural positions:

[0037] Where x and y represent the horizontal and vertical coordinates of a pixel in the nth frame of the infrared image to be registered, respectively; and These represent the horizontal and vertical coordinates of the pixel after spatial registration, respectively. This represents the affine transformation matrix obtained from at least three non-collinear stable structural positions.

[0038] After registration is completed, if the average registration error of the stable structure position is no higher than 1px, the corresponding image is retained; if it is higher than 1px, feature matching is re-executed; if the re-matching still does not meet the requirements, the corresponding image is discarded. If there are no more than 3 invalid images in a set of azimuth infrared time-series images, and there are no two consecutive invalid images, linear interpolation is performed using adjacent valid images; if there are 4 invalid images, or two consecutive invalid images, infrared images under the corresponding preset thermal excitation azimuth are re-acquired.

[0039] After quality screening and spatial registration are completed, an orientation thermal response image sequence is generated, and the pixel coordinate range of the detection unit is output simultaneously as the input of S210.

[0040] S2 specifically includes the following sub-steps: S210 and the edge computing node read the azimuth thermal response image sequence, the pixel coordinate range of the detection unit, and the detection configuration data output by S130, and process the infrared images group by group according to the preset thermal excitation orientation. Before formal testing, parameter calibration is completed using a healthy sample with the same coating system, construction thickness, curing conditions, and metal substrate type as the anti-corrosion layer to be tested, as well as a prefabricated interface microcrack sample.

[0041] The crack location, crack length, crack extension direction, and degree of closure in the prefabricated interface microcrack template are derived from the sample preparation record. The specific preparation method for the prefabricated interface microcrack template can be as follows: a release agent is coated or a polytetrafluoroethylene film with a thickness of no more than 10 μm is attached to a predetermined position on the surface of a metal substrate to simulate an unbonded interface; subsequently, a solvent-free polyester coating is sprayed and cured to form a calibration template with known closed-type interface microcrack characteristics.

[0042] The edge computing node writes the pixel neighborhood size, candidate delay frame range, local saddle point second-order change threshold, local saddle point gradient magnitude threshold, preset search radius, matching cost weight, and matching cost threshold into the calibration parameter file.

[0043] A pixel neighborhood refers to a set of consecutive pixels centered on the pixel to be analyzed, used to reduce the impact of temperature measurement noise from a single pixel on the image recognition result. This embodiment uses a 5px × 5px pixel neighborhood; when the pixel to be analyzed is located at the edge of the detection area, only valid pixels falling within the detection area are retained. For the m-th preset thermal excitation orientation, the u-th pixel neighborhood, and the k-th image frame acquisition time, the average temperature value of the pixel neighborhood is calculated according to the following formula:

[0044] in, This represents the average temperature value of the pixel's neighborhood. This represents the set of valid pixels contained in the neighborhood of the u-th pixel; Indicates the number of valid pixels; This represents the temperature value of the pixel with pixel coordinates (x, y) at the time of image frame acquisition in the kth frame; u represents the pixel neighborhood number; and k represents the image frame acquisition time number. The temperature value is directly output by the infrared imaging component; when the infrared imaging component only outputs grayscale images, the edge computing node reads the device's factory calibration table and completes the conversion according to the correspondence between grayscale values ​​and temperature values.

[0045] The average temperature values ​​of the same pixel neighborhood are arranged according to the image frame acquisition time to form a temperature response curve; the output power of the thermal excitation unit recorded by S120 is arranged according to a unified timestamp to form a thermal excitation reference sequence. The thermal phase value is the normalized phase value obtained by converting the temperature response curve relative to the thermal excitation reference sequence, and is used to characterize the degree of delay in heat transfer to the pixel neighborhood. The response delay frame number and thermal phase value are calculated according to the following formula:

[0046] in, Indicates the response latency in frames; Indicates the candidate delay frame number; This represents the thermally excited reference sequence after subtracting the average sequence value; This represents the sequence of temperature changes in the pixel neighborhood after subtracting the sequence average. Indicates the thermal phase value; Indicates a fixed sampling interval; This indicates the length of the analysis period corresponding to one preset thermal excitation orientation; This represents pi; argmax represents the number of candidate delay frames with the highest correlation. The range of candidate delay frames is derived from the template calibration results.

[0047] For example, with a fixed sampling interval of 0.1s, an analysis period of 10s, and a neighborhood delay of 4 frames for a certain pixel, the thermal phase value is 0.08. Arrange the thermal phase values ​​corresponding to the neighborhood of each pixel according to the pixel coordinates to form a thermal phase map; associate and save the thermal phase maps corresponding to each preset thermal excitation orientation to form a thermal phase map set.

[0048] S220, the edge computing node performs a 3px × 3px median filter on each thermal phase map, and then calculates the thermal phase gradient vector for pixel coordinates (x, y) according to the following formula: ; in, This represents the thermal phase gradient vector of the m-th preset thermal excitation orientation pixel with coordinates (x, y); This represents the thermal phase value corresponding to the pixel coordinates. The thermal phase gradient vectors are arranged according to the pixel coordinates to form a thermal phase gradient vector field.

[0049] A local saddle point is a pixel location in the thermal phase map that exhibits opposite second-order changes along the horizontal and vertical directions, respectively, and whose thermal phase gradient vector magnitude does not exceed the local saddle point gradient magnitude threshold. The second-order thermal phase changes of candidate pixel locations along the horizontal and vertical directions are calculated using the following formulas:

[0050]

[0051] in, and These represent the second-order thermal phase changes in the transverse and longitudinal directions, respectively.

[0052] When two second-order thermal phase changes have opposite signs, their absolute values ​​are both not lower than the local saddle point second-order change threshold, and the magnitude of the thermal phase gradient vector is not higher than the local saddle point gradient magnitude threshold, the candidate pixel position is identified as a local saddle point. Pixels whose positions cannot be simultaneously obtained on both sides are not included in the local saddle point identification. The threshold is derived from the calibration results of healthy templates and prefabricated interface microcrack templates. The edge computing node stores the pixel coordinates, detection unit number, preset thermal excitation orientation, thermal phase value, and thermal phase gradient vector of the local saddle point, forming a local saddle point record.

[0053] S230. According to the arrangement order of the preset thermal excitation orientations, for the local saddle point to be connected under the previous preset thermal excitation orientation, search for candidate local saddle points located within the preset search radius under the next preset thermal excitation orientation, and calculate the matching cost according to the following formula:

[0054] in, This represents the matching cost between the local saddle point a to be connected and the candidate local saddle point b. Indicates the distance to normalized pixel coordinates; This represents the normalized directional difference obtained by dividing the difference in the thermal phase gradient vector direction by 180°. This represents the adjacency constraint term for the detection unit. It is 0 when two local saddle points are located in the same or adjacent detection units, and 1 otherwise. , and These represent the weights corresponding to the three features.

[0055] Normalized pixel coordinate distance is calculated using the following formula:

[0056] in, , , and These represent the pixel coordinates of the two local saddle points; This represents the preset search radius, in pixels (px). The preset search radius, the three weights, and the matching cost threshold are all derived from the template calibration results.

[0057] When multiple candidate local saddle points exist, the candidate local saddle point with the lowest matching cost and not exceeding the matching cost threshold is selected to establish a one-to-one correspondence. Supplementary connections are allowed across one missing orientation; if no corresponding local saddle point is identified in two consecutive orientations, the current trajectory is terminated. Ordered local saddle points supported by at least four preset thermal excitation orientations are sequentially connected to form a saddle point migration trajectory. The migration distance, migration direction, trajectory continuity, orientation response stability, number of orientations supported by the trajectory, and the range of pixel coordinates covered by the trajectory are extracted to form a saddle point migration feature set.

[0058] Trajectory continuity is the ratio of the number of successfully connected adjacent azimuths to the total number of adjacent azimuths that can be connected; azimuth response stability is the ratio of the number of connected segments whose migration direction difference is not higher than a preset direction consistency threshold to the total number of all valid connected segments. The saddle point migration trajectory and saddle point migration feature set are inputs to S310 and are called by S510.

[0059] As a specific and feasible scenario, for a solvent-free polyester coating anti-corrosion layer with a thickness of 1 mm, the calibration parameters can be set as follows: the candidate delay frame range is 10 to 30 frames, and the local saddle point second-order variation threshold is 0.02. The local saddle point gradient magnitude threshold is 0.01. The preset search radius is 10px, and the matching cost weight is... , and The values ​​are set to 0.4, 0.4, and 0.2 respectively, with a matching cost threshold of 0.6.

[0060] S3 specifically includes the following sub-steps: S310: The edge computing node reads the saddle point migration trajectory and saddle point migration feature set output by S230. When the migration distance of a saddle point migration trajectory is not less than a preset migration distance threshold, the trajectory continuity is not less than a preset continuity threshold, the azimuth response stability is not less than a preset stability threshold, the trajectory supports at least 4 azimuths, and the number of consecutively missing preset thermal excitation azimuths does not exceed 1, the saddle point migration trajectory is determined as a valid saddle point migration trajectory.

[0061] The aforementioned thresholds are derived from the calibration results of the healthy template and the prefabricated interface microcrack template. For example, when the trajectory migration distance in the healthy template is usually less than 3px, and the trajectory migration distance in the prefabricated interface microcrack template is usually not less than 6px, the preset migration distance threshold can be set to 5px.

[0062] For the i-th valid saddle point migration trajectory, expand outward by a preset number of pixels based on the pixel coordinate range covered by the trajectory, forming a suspected interface micro-crack region:

[0063] in, This represents the suspected interface microcrack region corresponding to the i-th valid saddle point migration trajectory. and These represent the minimum and maximum horizontal coordinates within the pixel coordinate range covered by the trajectory, respectively; and These represent the minimum and maximum vertical coordinates, respectively; p represents the preset number of extended pixels in pixels (px); and i represents the effective saddle point migration trajectory number.

[0064] The preset number of extended pixels is derived from the calibration results of the abnormal thermal response range in the prefabricated interface microcrack template. For example, if the trajectory coverage pixel coordinate range can cover all abnormal thermal response pixels after being extended outward by 8px, the preset number of extended pixels is set to 8px.

[0065] A single interface microcrack may generate multiple effective saddle point migration trajectories that are close to each other. For any two suspected interface microcrack regions, the region overlap ratio is calculated using the following formula:

[0066] in, This represents the overlap ratio between the i-th suspected interface microcrack region and the j-th suspected interface microcrack region. Indicates the number of pixels contained in the overlapping portion; This indicates the number of pixels included after merging; j represents the migration trajectory number of another valid saddle point.

[0067] When the overlap ratio of the regions is not lower than the preset merging threshold, the two regions are merged, and the corresponding effective saddle point migration trajectories are jointly written into the suspected interface microcrack region record. The suspected interface microcrack region record includes at least the region number, pixel coordinate range, detection unit number, effective saddle point migration trajectory number, and corresponding saddle point migration feature set.

[0068] S320 and the edge computing node select thermal excitation units for performing bidirectional thermal scanning based on the records of suspected interface microcrack areas and the effective thermal excitation coverage area saved in S110. If the effective thermal excitation coverage area of ​​one thermal excitation unit can completely cover the suspected interface microcrack area, the thermal excitation unit with the smallest distance between the center of the coverage area and the center of the area is selected; if no single thermal excitation unit can completely cover the area, the minimum number of thermal excitation units that can collectively cover the area are selected, and the scan is performed sequentially in order of distance from nearest to farthest.

[0069] Two-way thermal scanning refers to targeting the same suspected interfacial microcrack region by progressively increasing the heat input per unit area at multiple thermal excitation levels to form a heating path, and then progressively decreasing the heat input per unit area at the opposite level to form a cooling path. For the h-th thermal excitation level, the heat input per unit area is determined by the following formula:

[0070] in, This represents the heat input per unit area corresponding to the h-th thermal excitation level, in J / cm². This represents the initial heat input per unit area; This represents the increment of heat input per unit area between two adjacent thermal excitation levels; h represents the thermal excitation level number.

[0071] The number of thermal excitation levels is denoted as L. The heating path follows... The cooling path is executed in the following order. The execution order must satisfy the following conditions:

[0072] in, and Consistent with the definition in S120.

[0073] The initial heat input per unit area, the incremental heat input per unit area, the number of thermal excitation levels, and the holding time of each thermal excitation level are all derived from the sample calibration results. For example, when the safety range is 0.6 J / cm² to 1.4 J / cm² and the incremental heat input per unit area is 0.2 J / cm², five thermal excitation levels can be set. Each thermal excitation level is held for 1 second, and the infrared imaging component continuously acquires images at a fixed sampling interval of 0.1 seconds.

[0074] S330. Before performing bidirectional thermal scanning, the edge computing node continuously reads the average temperature of suspected interface microcrack areas. When the difference between the average temperature of the area and the average temperature of the area calculated based on the reference infrared image is no higher than 0.3℃, 10 frames of local infrared images are continuously acquired at a fixed sampling interval of 0.1s to form a baseline local image sequence before scanning. The baseline local image sequence before scanning is used to calculate the baseline average temperature of each pixel's neighborhood before the start of bidirectional thermal scanning.

[0075] During bidirectional thermal scanning, the infrared imaging component continuously acquires local infrared images during each thermal excitation level holding period. Each image frame saves the suspected interface microcrack region number, thermal excitation unit number, heating or cooling path marker, thermal excitation level number, heat input per unit area, relative sampling time during level holding period, and image frame number.

[0076] Edge computing nodes reuse the quality screening rules and affine registration method in S130 to perform quality screening and spatial registration on the baseline local image before scanning and the local images acquired during bidirectional thermal scanning, and use the pixel coordinate range determined in S310 as the fixed local image cropping boundary. For the same suspected interface microcrack region, the heating path and cooling path use the exact same cropping boundary.

[0077] When there are no more than one invalid local image under a single thermal excitation level, linear interpolation is performed using adjacent valid images. When there are two or more invalid local images, or two consecutive invalid local images, a bidirectional thermal scanning infrared image corresponding to the suspected interface microcrack area is reacquired. Edge computing nodes establish a correspondence between heating path local images and cooling path local images with the same thermal excitation level number and the same relative sampling time, generating a bidirectional thermal scanning infrared image sequence.

[0078] The pre-scan baseline local image sequence, bidirectional thermal scan control sequence, and bidirectional thermal scan infrared image sequence are used as inputs to S410.

[0079] S4 specifically includes the following sub-steps: S410 and the edge computing node read the pre-scan baseline local image sequence, bidirectional thermal scan control sequence and bidirectional thermal scan infrared image sequence output by S330, and establish a correspondence according to the suspected interface microcrack area number, thermal scan path number, thermal excitation level number and relative sampling time.

[0080] A thermal scan path number of 1 indicates a heating path, and a value of 2 indicates a cooling path. The time interval during which one thermal excitation level is continuously executed is defined as the level holding period; consecutive image frames are selected from the end of the level holding period to form a stable sampling window. The number of image frames included in the stable sampling window is derived from the template calibration results. In this embodiment, each level holding period is 1 second, the fixed sampling interval is 0.1 seconds, and the last 3 frames of the level holding period are selected as the stable sampling window.

[0081] For the v-th thermal scan path, the u-th pixel neighborhood, and the h-th thermal excitation level, the local thermal phase response is calculated according to the following formula:

[0082] in, This indicates the local thermal phase response; v indicates the thermal scan path number; u has the same meaning as the pixel neighborhood number in S210; h has the same meaning as the thermal excitation level number in S320. This represents the time, in seconds, required for the absolute value of the change in the average temperature of a pixel's neighborhood between adjacent frames to reach its maximum value after the corresponding thermal excitation level has been executed. This represents the holding time for the h-th thermal excitation level, in seconds. It represents pi (π).

[0083] For example, if the grade retention time is 1 second, and a pixel's neighborhood experiences the largest adjacent frame temperature change 0.3 seconds after the grade begins, then the local thermal phase response is 0.6. .

[0084] For the same pixel neighborhood, the local thermal amplitude response and local gradient diffusion response are calculated according to the following formula:

[0085] in, Indicates the local thermal amplitude response; This represents the average temperature value of the pixel neighborhood within the stable sampling window; This represents the average temperature of the baseline before scanning, calculated from the local image sequence of the baseline before scanning. This represents the local gradient diffusion response; This represents the set of neighboring pixels that actually exist around the u-th pixel. This represents the number of neighboring pixels; w represents the neighboring pixel number. The heating path response and cooling path response under the same thermal excitation level are associated and saved to form a bidirectional local thermal response pair.

[0086] S420, local thermal phase response, local thermal amplitude response, and local gradient diffusion response have different dimensions. Edge computing nodes read the calibration parameter file and normalize the three types of local thermal responses according to the following formula:

[0087] in, represents the normalized response value; z represents the response value to be normalized. and These represent the minimum and maximum calibration values ​​determined based on the healthy template and the precast interface microcrack template, respectively. When the normalization result is zero, the normalization result is set to 0. The upper and lower limits of normalization do not use the maximum and minimum values ​​of the current suspected interface microcrack region itself, so as to ensure that the calculation results between different regions can be compared.

[0088] For the v-th thermal scan path, the u-th pixel neighborhood, and the h-th thermal excitation level, the comprehensive thermal response is calculated according to the following formula:

[0089] in, Indicates the overall thermal response; , and These represent the normalized local thermal phase response, local thermal amplitude response, and local gradient diffusion response, respectively. , and These represent the weights corresponding to the three types of local thermal responses, and The three weights are derived from the sample calibration results. During calibration, the candidate weight combinations are traversed with a fixed step size of 0.05, and the weight combination that maximizes the difference in overall thermal response between the healthy sample and the prefabricated interface microcrack sample is selected.

[0090] For example, in a specific implementation scenario, the local thermal phase response weights selected using the above calibration method... The local thermal amplitude response weight is 0.40. The local gradient diffusion response weight is 0.35. It is 0.25.

[0091] Using the heat input per unit area in S320 as the horizontal axis and the comprehensive thermal response as the vertical axis, connect the data points corresponding to the heating path in order of thermal excitation level to form the heating response path; connect the data points corresponding to the cooling path in reverse order to form the cooling response path; associate the two paths to obtain the thermal hysteresis loop. In the closed-interface microcrack region, the contact state of the crack wall changes under the heating and cooling paths, causing a non-overlapping region between the two paths.

[0092] S430. For the u-th pixel neighborhood, calculate the loop gap feature, phase transition feature, and path asymmetry feature according to the following formula:

[0093]

[0094]

[0095] in, This represents the hysteresis loop gap feature, used to characterize the discrete area between the heating response path and the cooling response path; This indicates the phase transition characteristic, used to characterize the largest abrupt change in heat transfer delay between adjacent thermal excitation levels; This indicates the asymmetric feature of the path, used to characterize the normalized difference between two paths; The meaning of the incremental heat input per unit area is consistent with that in S320; the meaning of L is consistent with that of the number of thermal excitation levels in S320. This is a positive number set to prevent the denominator from being 0. In this embodiment, we take... . and These represent the combined thermal responses under the heating path (v=1) and the cooling path (v=2), respectively.

[0096] For multiple effective saddle point migration trajectories associated with the same suspected interface microcrack region, the trajectory with the most supported orientations is selected as the main effective saddle point migration trajectory; when the number of supported orientations is the same, the trajectory with the highest orientation response stability is selected. Along the extension direction of the main effective saddle point migration trajectory, continuous pixel neighborhoods that simultaneously satisfy the loop gap threshold, phase transition threshold, and path asymmetry threshold are statistically analyzed to form continuous abnormal segments. The three thresholds and the minimum number of continuous pixel neighborhoods are all derived from the sample calibration results.

[0097] For example, if the migration trajectory of the main effective saddle point crosses 18 pixel neighborhoods, and 14 consecutive pixel neighborhoods simultaneously meet all three thresholds, the corresponding continuous abnormal segment is retained; if only one isolated pixel neighborhood meets the thresholds, it is marked as an isolated anomaly. The suspected interface microcrack region number, pixel neighborhood number, three types of local thermal response, comprehensive thermal response, three types of anomaly features, coordinates of continuous abnormal pixel neighborhoods, and continuous distribution length are associated and saved to form a thermal hysteresis feature set, which is then input into S510.

[0098] S5 specifically includes the following sub-steps: S510 and the edge computing node read the saddle point migration feature set output by S230 and the thermal hysteresis feature set output by S430, and establish a corresponding relationship according to the suspected interface microcrack region number. For each suspected interface microcrack region, the main effective saddle point migration trajectory and continuous abnormal segments are read. Continuous abnormal segments refer to the set of pixel neighborhoods that are continuously distributed along the extension direction of the main effective saddle point migration trajectory and simultaneously satisfy the loop gap threshold, phase transition threshold, and path asymmetry threshold.

[0099] Around the migration trajectory of the main effective saddle point corresponding to the r-th suspected interface microcrack region, a preset matching width is extended to both sides of the trajectory to form a trajectory strip region. The trajectory strip region is used to accommodate the allowable deviation between the local saddle point position and the actual crack center position. The preset matching width is derived from the calibration results of the prefabricated interface microcrack template. The spatial overlap ratio between the trajectory strip region and the continuous abnormal section is calculated according to the following formula:

[0100] in, This represents the spatial overlap ratio corresponding to the r-th suspected interface microcrack region; Indicates a band-shaped region of trajectory; Indicates a continuous abnormal section; Indicates the number of pixels contained in the overlapping portion; The number of pixels included after merging is indicated; r represents the number of the suspected micro-crack area on the interface.

[0101] For example, if the trajectory strip area contains 120px, the continuous abnormal segment contains 100px, and the overlapping part contains 80px, then the spatial overlap ratio is 0.57; when the preset spatial overlap threshold is 0.50, the corresponding area is retained.

[0102] When the spatial overlap ratio is not lower than the preset spatial overlap threshold, the migration distance, trajectory continuity, and orientation response stability of the main effective saddle point migration trajectory reach the corresponding thresholds adopted in S310, and the continuous distribution length of the continuous abnormal section is not lower than the preset minimum continuous distribution length, the suspected interface microcrack region is identified as a candidate region for closed interface microcracks. This judgment can exclude interface bubbles that only exhibit heat flow circulation characteristics but do not show changes in the crack wall contact state.

[0103] When two candidate regions for closed-type interface microcracks have pixel overlap, the two regions are merged. If there is no pixel overlap, and the distance between adjacent endpoints of two consecutive abnormal segments is not higher than a preset endpoint distance threshold, and the directional difference between the two main effective saddle point migration trajectories is not higher than a preset directional difference threshold, the merging is also performed. The preset spatial overlap threshold, preset minimum continuous distribution length, preset endpoint distance threshold, and preset directional difference threshold are all derived from the template calibration results. The edge computing node stores the candidate region numbers and corresponding main effective saddle point migration trajectory numbers before and after merging.

[0104] S520. For each merged closed-type interface microcrack candidate region, the edge computing node uses an image recognition algorithm to generate a binary image of the candidate region: the pixel positions belonging to the continuous abnormal segment are assigned a value of 1, and the other pixel positions are assigned a value of 0. Adjacent abnormal pixels are connected using the 8-neighborhood connectivity rule, and centerline thinning processing is performed on the connected regions to obtain a candidate centerline with a width of 1px.

[0105] When the candidate centerline contains multiple branches, the degree of overlap between each branch and the migration trajectory of the main effective saddle point is calculated. The branch with the highest degree of overlap and the longest continuous length is retained as the centerline of the closed interface microcrack, hereinafter referred to as the crack centerline.

[0106] The edge computing node calls the pixel coordinate to actual coordinate transformation matrix stored in S110 to convert the pixel coordinates of the crack centerline into the actual coordinates of the surface of the anti-corrosion layer to be inspected.

[0107] Where X and Y represent the actual transverse and longitudinal coordinates of the surface of the anti-corrosion layer to be tested, respectively, in mm; x and y represent the transverse and longitudinal coordinates of the crack centerline pixel, respectively. This represents the transformation matrix from pixel coordinates to actual coordinates for the c-th detection unit; c represents the detection unit number.

[0108] For bent pipes or irregularly shaped components, local transformation matrices are called separately for each detection unit, without using a uniform conversion ratio for the entire image. For all actual coordinate points along the crack centerline, least-squares linear fitting is used to obtain the main extension direction. When the linear fitting residual exceeds a preset linear fitting residual threshold, the crack centerline is divided into multiple centerline segments according to a preset segment length, and the local extension direction is saved for each segment. The preset linear fitting residual threshold and preset segment length are derived from the template calibration results.

[0109] For the r-th candidate region of closed-type interface microcracks, the average regional anomaly characteristics corresponding to the continuous anomaly segments are calculated respectively:

[0110] in, , and These represent the regional average loop gap characteristics, the regional average phase transition characteristics, and the regional average path asymmetry characteristics, respectively. This indicates the number of pixel neighborhoods contained in a continuous abnormal segment; u represents the pixel neighborhood number. After completing the region averaging calculation, the corresponding normalized features are obtained according to the normalization method in S420.

[0111] The risk score is calculated using the following formula:

[0112] in, Indicates risk score; This represents the normalized migration distance; Indicates trajectory continuity; Indicates the stability of the azimuth response; , and These represent the three normalized regional average anomaly features; This represents the length of the normalized continuous distribution. Indicates the proportion of overlap in space; to These represent the risk weights of the corresponding features, and the sum of the weights is 1.

[0113] The normalization upper and lower limits and risk weights are derived from the calibration results of prefabricated interface microcrack samples with different crack lengths, degrees of closure, and burial locations. During calibration, candidate risk weight combinations are traversed using a fixed step size of 0.05, and the weight combination that best matches the actual crack state is selected based on the cross-sectional verification results. As a specific and feasible scenario, the risk weights in the above formula... to The possible values ​​are: 0.10, 0.15, 0.10, 0.20, 0.15, 0.10, 0.10, and 0.10.

[0114] The risk level is determined using the following formula:

[0115] in, Indicates the risk level; This represents the risk scoring threshold between low-risk and medium-risk levels. This represents the risk scoring threshold between medium and high risk levels. The two risk scoring thresholds are derived from the cross-sectional verification results of prefabricated interface microcrack samples. For example, 0.45 When the risk score is 0.70, the candidate area with a risk score of 0.76 is identified as a high-risk area.

[0116] The S530 edge computing node overlays the crack centerline, the actual coordinates of the crack centerline, the main extension direction, and the risk level onto the reference infrared image acquired by the S110, generating a closed-type interface microcrack image recognition report.

[0117] The image recognition report for closed-type interface microcracks should include at least the following: the region to be detected, the detection unit, the candidate region for the closed-type interface microcrack, the candidate region before merging, the migration trajectory number of the main effective saddle point, the pixel coordinates of the crack centerline, the actual coordinates of the crack centerline, the main extension direction, the local extension direction, the migration distance, the trajectory continuity, the azimuth response stability, the spatial overlap ratio, the regional average loop gap characteristics, the regional average phase transition characteristics, the regional average path asymmetry characteristics, the continuous distribution length, the risk score, the risk level, the detection configuration data version, and the calibration parameter version.

[0118] When none of the suspected interface microcrack areas meet the candidate area generation conditions in S510, the edge computing node outputs the image recognition result of "no closed interface microcracks detected" and saves the detection area number, detection configuration data version and calibration parameter version that have been detected.

[0119] When the azimuth thermal response image sequence or bidirectional thermal scanning infrared image sequence is incomplete, the image spatial registration error exceeds the threshold, the thermal excitation parameters are outside the safe range, or the calibration parameter version does not match the coating system of the anti-corrosion layer to be inspected, the edge computing node outputs a "detection abnormal, re-acquisition required" status, without outputting a no-detection conclusion. Finally, the closed-type interface microcrack image recognition report is sent to the inspection terminal or maintenance management system for scheduling review and maintenance.

[0120] Example 2: Figures 2 to 4 As shown, this embodiment provides an intelligent sensing system for detecting microcracks in solvent-free polyester coating anti-corrosion layers, including multiple thermal excitation units, an infrared imaging component, and an edge computing node. The edge computing node includes: The azimuth thermal response acquisition module is used to divide the anti-corrosion layer to be tested into detection units, control the thermal excitation unit to apply thermal excitation sequentially along multiple preset thermal excitation azimuths, and control the infrared imaging component to acquire and register infrared time-series images to generate an azimuth thermal response image sequence. The saddle point migration analysis module is used to extract the thermal phase values ​​of the pixel neighborhood based on the azimuth thermal response image sequence, construct a thermal phase gradient vector field, identify and connect local saddle points, and form a saddle point migration trajectory and a saddle point migration feature set. The bidirectional thermal scanning module is used to filter valid saddle point migration trajectories based on the saddle point migration feature set, determine suspected interface microcrack regions, and perform bidirectional thermal scanning including heating and cooling paths. The thermal hysteresis feature extraction module is used to construct thermal hysteresis loops based on bidirectional thermal scanning infrared image sequences, extract loop gap features, phase transition features, and path asymmetry features to form a thermal hysteresis feature set. The crack identification output module is used to spatially fuse the saddle point migration feature set with the thermal hysteresis feature set to determine the candidate region of closed interface microcracks, extract the center line of the closed interface microcracks, and output the location, extension direction and risk level.

[0121] All the above formulas are performed using dimensionless numerical calculations; the relevant formulas are based on empirical models that approximate the real situation, obtained through extensive data collection and software simulation fitting. The preset parameters and thresholds involved in the formulas can be conventionally set and adjusted by those skilled in the art according to the physical constraints of the actual application scenario.

[0122] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0123] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0124] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting microcracks in solvent-free polyester coating anticorrosion layers, characterized in that, Includes the following steps: S1. Divide the anti-corrosion layer to be tested into detection units, apply thermal excitation sequentially along multiple preset thermal excitation directions, collect and register the corresponding infrared time sequence images, and generate a sequence of directional thermal response images. S2. Based on the azimuth thermal response image sequence, extract the thermal phase value of the pixel neighborhood, construct the thermal phase gradient vector field, identify local saddle points, and connect the corresponding local saddle points according to the preset thermal excitation azimuth order to form the saddle point migration trajectory and saddle point migration feature set. S3. Select effective saddle point migration trajectories based on the saddle point migration feature set, determine the suspected interface microcrack region around the effective saddle point migration trajectory, and perform bidirectional thermal scanning including heating path and cooling path on the suspected interface microcrack region to acquire bidirectional thermal scanning infrared image sequence. S4. Based on the bidirectional thermal scanning infrared image sequence, extract the local thermal phase response, local thermal amplitude response and local gradient diffusion response, construct a thermal hysteresis loop, extract the loop gap feature, phase transition feature and path asymmetry feature to form a thermal hysteresis feature set. S5. Spatially fuse the saddle point migration feature set and the thermal hysteresis feature set to determine the candidate region of closed interface microcracks, extract the center line of closed interface microcracks, and output the location, extension direction and risk level of closed interface microcracks.

2. The method for detecting microcracks in solvent-free polyester coating anticorrosion layers according to claim 1, characterized in that, S1 specifically includes: Select the area to be tested on the surface of the anti-corrosion layer to be tested, divide the area to be tested into multiple detection units, set up multiple thermal excitation units, infrared imaging components and edge computing nodes, and establish detection configuration data between detection units, preset thermal excitation orientation, stable structure position and pixel coordinates. Based on the detection configuration data, the corresponding thermal excitation units are controlled sequentially according to the preset thermal excitation orientation to apply thermal excitation within a safe range, and infrared images are acquired simultaneously to form the original orientation thermal response image sequence; The original azimuth thermal response image sequence is subjected to quality screening and spatial registration to generate an azimuth thermal response image sequence, and the pixel coordinate range of the detection unit is saved.

3. The method for detecting microcracks in solvent-free polyester coating anticorrosion layers according to claim 1, characterized in that, S2 specifically includes: Read the azimuth thermal response image sequence, extract the temperature response curve of the pixel neighborhood for the preset thermal excitation azimuth, and calculate the thermal phase value based on the response delay of the temperature response curve relative to the thermal excitation reference sequence to form a thermal phase map set. The thermal phase map set is filtered, the thermal phase gradient vector is calculated, the thermal phase gradient vector field is constructed, and local saddle points are identified based on the horizontal and vertical second-order thermal phase changes, forming local saddle point records.

4. The method for detecting microcracks in solvent-free polyester coating anticorrosion layers according to claim 3, characterized in that, Also includes: According to the preset thermal excitation orientation sequence, local saddle points in adjacent orientations are matched and connected to form saddle point migration trajectories. The migration distance, migration direction, trajectory continuity, orientation response stability, and trajectory coverage pixel coordinate range are extracted to form a saddle point migration feature set.

5. The method for detecting microcracks in solvent-free polyester coating anticorrosion layers according to claim 1, characterized in that, S3 specifically includes: Read the saddle point migration trajectory and saddle point migration feature set, filter the effective saddle point migration trajectory, expand and merge the overlapping areas around the effective saddle point migration trajectory to form a record of suspected interface microcrack areas. Based on the records of suspected interface microcrack areas, the corresponding thermal excitation unit is selected, and the heat input per unit area is increased and decreased sequentially according to the thermal excitation level to form a heating path and a cooling path. Before the start of bidirectional thermal scanning, a baseline local image sequence is acquired. During the bidirectional thermal scanning process, local infrared images are acquired, filtered, registered, and cropped. The heating path images and cooling path images corresponding to the same thermal excitation level and relative sampling time are paired to form a bidirectional thermal scanning infrared image sequence.

6. The method for detecting microcracks in solvent-free polyester coating anticorrosion layers according to claim 1, characterized in that, S4 specifically includes: Read the baseline local image sequence and the bidirectional thermal scan infrared image sequence before scanning. According to the thermal scan path, thermal excitation level and pixel neighborhood, extract the local thermal phase response, local thermal amplitude response and local gradient diffusion response to form a bidirectional local thermal response pair. The bidirectional local thermal response pairs are normalized and weighted to obtain a comprehensive thermal response. The comprehensive thermal responses corresponding to the heating and cooling paths are connected according to the heat input per unit area to construct a thermal hysteresis loop.

7. The method for detecting microcracks in solvent-free polyester coating anticorrosion layers according to claim 6, characterized in that, Also includes: Based on the thermal hysteresis loop, the loop gap features, phase transition features, and path asymmetry features are extracted, and the neighborhood of continuous abnormal pixels is statistically analyzed along the migration trajectory of the main effective saddle point to form a continuous abnormal segment and thermal hysteresis feature set.

8. The method for detecting microcracks in solvent-free polyester coating anticorrosion layers according to claim 1, characterized in that, S5 specifically includes: Read the saddle point migration feature set and thermal hysteresis feature set, establish a correspondence according to the suspected interface microcrack region number, calculate the spatial overlap ratio between the trajectory band region and the continuous abnormal section, and generate and merge closed interface microcrack candidate regions. The candidate regions of closed interface microcracks are connected and their centerlines are refined. The centerlines of the closed interface microcracks are extracted, their actual locations are calculated, their extension directions are determined, and the risk level is determined based on the saddle point migration feature set, thermal hysteresis feature set, and spatial overlap ratio.

9. The method for detecting microcracks in solvent-free polyester coating anticorrosion layers according to claim 8, characterized in that, Also includes: The center line, actual location, extension direction, and risk level of the closed interface microcrack are superimposed onto the baseline infrared image to generate a closed interface microcrack image recognition report.

10. An intelligent sensing system for detecting microcracks in solvent-free polyester coating anticorrosion layers, employing the image detection method for solvent-free polyester coating anticorrosion layers as described in any one of claims 1 to 9, characterized in that, It includes multiple thermal excitation units, infrared imaging components, and edge computing nodes.

Images