Membrane material surface defect detection method and system based on machine vision

Abstract

The invention discloses a film material surface defect detection method and system based on machine vision, and relates to the technical field of machine vision. Joint modeling is carried out by utilizing a local reflectivity difference value Ploc (x, y) of a pixel coordinate (x, y), a local stacking disturbance index S (x, y) of the pixel coordinate (x, y) and a local gray variance Wvar (x, y) of the pixel coordinate (x, y) in a film material image irradiated by a dual-angle light source, and the detection sensitivity of asymmetric optical disturbance caused by trace salt mist deposition is enhanced. Compared with the current situation that a traditional visual detection method is insufficient in response to early corrosion spots, the method has the advantages that a light reflection subsurface corrosion map can be effectively formed through quantitative combination of two factors of reflection anomaly and structural disturbance in the image, and rapid positioning and early warning of the membrane material salt spray corrosion initial stage are achieved.

Description

Technical Field

[0001] The present invention relates to the field of machine vision technology, and in particular to a method and system for detecting surface defects of film materials based on machine vision. Background Art

[0002] The present invention focuses on the evolution process of microscopic defects produced by membrane materials in a highly corrosive salt spray environment. Relying on a strategy that combines image processing and computational models, it extracts and evaluates multi-dimensional features such as reflective anomalies, morphological fluctuations, and structural evolution on the membrane surface. It is particularly suitable for environments such as wind power equipment, marine engineering equipment, and communication packaging units that have extremely high requirements for material surface integrity. It can predict the potential distribution of membrane corrosion points and achieve early intervention in structural defects.

[0003] Currently, offshore wind turbines are subject to harsh environments characterized by high salt levels, high humidity, strong winds, and drastic temperature swings year-round. Their communication and signal control units are encapsulated with transparent polyimide (PI) or PET protective films. Early micro-corrosion caused by salt spray on these membranes can lead to signal obstruction and reflection anomalies. Electrochemical corrosion at these sites can expand, eventually forming perforations. This prevents effective early detection, and traditional inspections often occur after the failure stage. However, traditional membrane inspection methods rely on manual visual inspection or regular illumination imaging for static defect detection. These methods are only capable of identifying late-stage defects such as larger perforations and cracks, while exhibiting minimal response to early-stage optical reflection anomalies, localized asymmetric disturbances, and micro-undulations in texture. In particular, they are unable to predict whether a corrosion pit will develop into a structural crack. Furthermore, existing machine vision systems are often limited to single-light sources or single-angle illumination, making them susceptible to pseudo-defects such as interfering light, dust particles, and water mist. This results in high misjudgment rates, delayed warnings, and a lack of intelligent perception of "defect evolution trends" in salt spray environments.

[0004] The fundamental problem with this situation is that current visual inspection technology fails to model and link the evolutionary chain of "reflective disturbances, structural anomalies, and crack perforations" in membrane materials in highly corrosive environments, lacking the ability to quantitatively identify and trend weak disturbance features in the image that have the potential to evolve. When early corrosion points such as initial salt crystal deposition, surface stress loosening, and micro-area wrinkling are just forming, the changes in the image are often only reflected in non-obvious features such as grayscale perturbations, texture bending, and slight symmetry loss, which are easily overlooked by traditional algorithms. If not identified and processed in a timely manner, the corrosion points will continue to expand into fine cracks and rapidly become unstable under wind pressure or vibration, eventually causing communication module failure or overall equipment degradation, posing a significant risk to equipment safety and operation and maintenance costs. Therefore, a high-sensitivity visual inspection solution that integrates light reflection anomaly analysis and structural stress evolution prediction is urgently needed to achieve early detection, proactive warning, and trend control of membrane defects in salt spray environments. Summary of the Invention

[0005] In view of the deficiencies in the prior art, the present invention provides a method and system for detecting surface defects of film materials based on machine vision, which solves the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: comprising the following steps: S1. Setting a visual sensor to collect image data of the outer sealing film of the signal control unit of the wind turbine equipment and transmitting the image data of the outer sealing film to the defect detection cloud server; S2. Preprocess the outer film sealing image data in the defect detection cloud server to obtain a standardized image set, and then perform feature extraction on the standardized image set to obtain a disturbance data set; S3. Calculate and output the light reflection disturbance anomaly index function Gpot based on the disturbance data set, conduct preliminary comparative evaluation, preliminarily judge the salt spray disturbance, and trigger the stress evolution trend inference mechanism; S4. After triggering the stress evolution trend inference mechanism, calculate and output the structural stress evolution tension function Jstress, and calculate and output the membrane material corrosion risk function Rtotal based on the structural stress evolution tension function Jstress; S5. Set the risk interval threshold, conduct a secondary comparative evaluation of the membrane material corrosion risk function Rtotal and the risk interval threshold, and classify the secondary comparative evaluation results and implement the corresponding response mechanism.

[0007] Preferably, said S1 includes S11 and S12; S11. At the outer sealing film of the communication and signal control unit of the seaside wind turbine equipment, a fixed dual-angle light source is simultaneously set to provide fill light for the outer sealing film, and a visual sensor is used to perform multi-frame sampling and shooting of the outer sealing film to obtain image data of the outer sealing film in real time; The fixed dual-angle light source includes a main light source and an auxiliary light source; the main light source is arranged at a 15° angle to emphasize specular reflection, and the auxiliary light source is arranged at a 45° angle to emphasize diffuse reflection and scattering information. At the same time, a 5600K white light LED light source is used in conjunction with a diffuse reflection background plate to eliminate external stray light; The multi-frame sampling shooting is performed by continuously shooting 3 frames of outer sealing film image data at the outer sealing film of each communication and signal control unit; S12. Through the communication module of the visual sensor, use the 5G wireless communication network to wirelessly integrate the visual sensor with the defect detection cloud server, and transmit the outer sealing film image data collected in real time to the defect detection cloud server.

[0008] Preferably, said S2 includes S21 and S22; S21. Receiving outer film sealing image data in real time in the defect detection cloud server, and preprocessing the outer film sealing image data to obtain a standardized image set; The preprocessing includes denoising, block processing, multi-frame fusion and grayscale and contrast normalization; The denoising is performed by bilateral filtering, Gaussian filtering and Laplace inversion to remove background low-frequency distribution while retaining edge information; Therefore, the block processing is performed by cutting the outer film image data into 64×64 blocks, and setting the lower left corner of the outer film image data to the pixel coordinate vertex (0,0). The horizontal and vertical axes of each block constitute the pixel coordinates (x, y), where x represents the horizontal pixel coordinate and y represents the vertical pixel coordinate. The grayscale and contrast normalization is performed on the outer sealing film image data processed in blocks by using histogram equalization, adaptive contrast optimization and Min-Max grayscale normalization; The multi-frame fusion simultaneously performs mean and temporal variance analysis on the pixels of the three frames of outer film image data collected continuously from the same viewing angle within the time window, and sets a floating artifact judgment threshold to eliminate the outer film image data with a value higher than the mean and temporal variance; S22. Perform feature extraction on the standardized image set using an image feature extraction technique to obtain a perturbation dataset, and perform Z-score standardization on the obtained perturbation dataset, subtracting the mean and dividing the standard deviation to obtain a standard normal distribution of the parameters in the perturbation dataset, thereby eliminating the dimensionality effect of the parameters in the perturbation dataset; The perturbation data set includes the local reflectivity difference △Ploc(x,y) of the pixel coordinate (x, y), the local stacking perturbation index S(x, y) of the pixel coordinate (x, y), the local grayscale variance Wvar(x, y) of the pixel coordinate (x, y), the local structure residual tensor field Tres(x, y) of the pixel coordinate (x, y), the local gradient standard deviation Jb(x, y) of the pixel coordinate (x, y) and the surface morphological curvature K(x, y) of the pixel coordinate (x, y).

[0009] Preferably, said S3 includes S31 and S32; S31. Constructing a light reflection disturbance risk algorithm model, extracting a disturbance data set and inputting it into the light reflection disturbance risk algorithm model, calculating and outputting a light reflection disturbance anomaly index function Gpot, and measuring the optical disturbance index present in the collected outer film image data; The light reflection disturbance anomaly index function Gpot is calculated and output by the following light reflection disturbance risk algorithm model; ; Where Gpot(x, y) represents the light reflection disturbance anomaly exponential function of the pixel coordinate (x, y), log represents the logarithmic function, represents the anti-divergence constant, which is set to 1×10 -5 To avoid denominators being zero, the logarithmic function log is used to calculate anomalies.

[0010] Preferably, S32, after obtaining the light reflection disturbance anomaly index function Gpot (x, y) of the pixel coordinates (x, y) of each block, a preliminary comparative evaluation is performed, wherein the preliminary comparative evaluation is performed by sampling a group of outer sealing film image data samples containing known defects, and calculating and outputting the light reflection disturbance anomaly index function Gpot (x, y) of the pixel coordinates (x, y), and selecting the light reflection disturbance anomaly index function Gpot (x, y) of the minimum pixel coordinates (x, y) as the disturbance threshold F1; The disturbance threshold F1 is compared with the light reflection disturbance anomaly index function Gpot(x, y) of the pixel coordinates (x, y) of all blocks to preliminarily determine the characteristics of salt spray erosion and trigger the stress evolution trend inference mechanism. The specific evaluation content is as follows; When the light reflection disturbance anomaly index function Gpot(x, y) of the pixel coordinate (x, y) is greater than the disturbance threshold F1, it indicates disturbance anomaly. It is preliminarily judged that salt spray disturbance exists. At this time, the proportion of disturbance anomalies in all the current outer film image data blocks is analyzed. If the proportion of disturbance anomalies exceeds 30%, the stress evolution trend inference mechanism is triggered; If the proportion of disturbance anomalies is less than or equal to 30%, the detection frequency will be increased by 50%; When the light reflection disturbance anomaly index function Gpot(x, y) of the pixel coordinate (x, y) is ≤ the disturbance threshold F1, it means that the disturbance is normal and does not belong to the corrosion risk area. At this time, the next block is traversed.

[0011] Preferably, said S4 includes S41 and S42; S41. After preliminary comparative evaluation of the trigger stress evolution trend inference mechanism, the disturbance data set is extracted and the structural stress evolution tension function Jstress is calculated and output to analyze the evolution of defects caused by salt spray erosion. The structural stress evolution tension function Jstress is calculated and outputted by the following algorithm formula; ; Where Jstress(x, y) represents the structural stress evolution tension function of the pixel coordinate (x, y), Represents the divergence of the tensor field.

[0012] Preferably, S42, based on the obtained structural stress evolution tension function Jstress and light reflection disturbance anomaly index function Gpot, a comprehensive calculation is performed to output a film corrosion risk function Rtotal, and a comprehensive analysis is performed on the comprehensive erosion and evolution trend risk in each outer film image data; The membrane corrosion risk function Rtotal is calculated and outputted by the following algorithm formula: ; Where A represents the total number of all blocks assessed as disturbance anomalies, L{Gopt>F1} represents the indicator function, a1 and a2 represent the preset weight values of the light reflection disturbance anomaly index function Gpot and the structural stress evolution tension function Jstress, respectively. The specific values are set by the user, and a1+a2=1.

[0013] Preferably, said S5 includes S51 and S52; S51, reversely setting a risk interval threshold based on the critical state of structural damage of the membrane material in the salt spray test, wherein the risk interval threshold includes an upper risk threshold Q1 and a lower risk threshold Q2; Among them, the upper risk threshold Q1 is the critical value between erosion and defects; the lower risk threshold Q2 is the critical value between normal and erosion; The real-time obtained membrane corrosion risk function Rtotal is compared and evaluated with the risk interval threshold to determine the overall membrane corrosion risk of the currently acquired outer film image data. The evaluation is then graded based on the secondary comparison and evaluation results. The specific evaluation contents are as follows; When the membrane material corrosion risk function Rtotal is less than the lower risk threshold Q2, the outer sealing membrane of the communication and signal control unit of the current seaside wind power equipment is classified as level one risk; When the lower risk threshold Q2 ≤ membrane corrosion risk function Rtotal < upper risk threshold Q1, the outer sealing membrane of the communication and signal control unit of the current seaside wind power equipment is classified as a secondary risk; When the membrane material corrosion risk function Rtotal ≥ the risk upper limit threshold Q1, the outer sealing membrane of the communication and signal control unit of the current seaside wind power equipment is divided into the third level of risk.

[0014] Preferably, S52, based on the secondary comparative evaluation, corresponding response measures are executed based on the classification of the outer sealing film of the communication and signal control unit of the current seaside wind power equipment. The specific response measures are as follows; When classified as level one risk, it means that the outer sealing film of the communication and signal control unit of the current seaside wind power equipment is blocked by salt fog. At this time, no management is required and it should be cleaned as planned; When it is classified as a Level 2 risk, it means that the outer sealing membrane of the communication and signal control unit of the current offshore wind turbine equipment is corroded by salt spray, but there are no structural defects. At this time, an early warning prompt is generated and the outer sealing membrane of the communication and signal control unit of the offshore wind turbine equipment is immediately cleaned of salt spray. When it is classified as the third level of risk, it means that the outer sealing film of the communication and signal control unit of the current seaside wind power equipment is corroded by salt spray, resulting in defects in the membrane material structure. At this time, the operation of the current seaside wind power equipment is stopped, and an early warning is generated to replace the outer sealing film of the communication and signal control unit.

[0015] A film surface defect detection system based on machine vision, including a vision acquisition module, a film data extraction module, a salt spray disturbance analysis module, a disturbance evolution trend analysis module and a defect response module; The visual acquisition module collects image data of the outer sealing film of the signal control unit of the wind power equipment by setting a visual sensor, and transmits the image data of the outer sealing film to the defect detection cloud server; The film material data extraction module pre-processes the outer film sealing image data in the defect detection cloud server to obtain a standardized image set, and then extracts features from the standardized image set to obtain a disturbance data set; The salt fog disturbance analysis module calculates and outputs the light reflection disturbance anomaly index function Gpot based on the disturbance data set, performs preliminary comparative evaluation, preliminarily determines the salt fog disturbance, and triggers the stress evolution trend inference mechanism; The disturbance evolution trend analysis module calculates and outputs the structural stress evolution tension function Jstress after triggering the stress evolution trend inference mechanism, and calculates and outputs the membrane material corrosion risk function Rtotal based on the structural stress evolution tension function Jstress; The defect response module sets a risk interval threshold, performs a secondary comparative evaluation on the film material corrosion risk function Rtotal and the risk interval threshold, and classifies the secondary comparative evaluation results and executes a corresponding response mechanism.

[0016] The present invention provides a method and system for detecting film surface defects based on machine vision. It has the following beneficial effects: (1) This method constructs a light reflection disturbance anomaly index function Gpot in S3 and jointly models the local reflectivity difference △Ploc(x,y) of the pixel coordinates (x,y) in the film image under dual-angle light source illumination, the local stacking disturbance index S(x,y) of the pixel coordinates (x,y), and the local grayscale variance Wvar(x,y) of the pixel coordinates (x,y) to enhance the detection sensitivity of asymmetric optical disturbances caused by trace salt spray deposition. Compared with the current situation that traditional visual detection methods are insufficient in responding to early corrosion points, the present invention can effectively form a light reflection corrosion map through the quantitative combination of the two factors of "reflection anomaly and structural disturbance" in the image, realize the rapid positioning and early warning of the "initial stage" of film salt spray corrosion, and significantly improve the foresight and completeness of defect identification.

[0017] (2) This method constructs a structural stress evolution tension function Jstress based on the local structural residual tensor field Tres(x,y) of the pixel coordinates (x,y) in the perturbation data set, the local gradient standard deviation Jb(x,y) of the pixel coordinates (x,y), and the surface morphological curvature K(x,y) of the pixel coordinates (x,y), to achieve a quantitative assessment of the evolutionary behaviors of potential crack propagation, edge instability, and geometric mutation in the local area of the membrane material. This mechanism can not only further confirm whether the latent corrosion caused by the highlight area of the light reflection perturbation abnormal exponential function Gpot has transformed into structural damage, but also focus on the defect source points with severe boundary perturbations and obvious curvature changes through the divergence enhancement factor. Compared with the limitations of static defect recognition in existing technologies, the present invention can achieve "development judgment" based on the texture tension evolution trend and has the ability of structural intelligent evolution reasoning.

[0018] (3) This method proposes a membrane corrosion risk function Rtotal in S5. By fusing the light reflection disturbance anomaly index function Gpot and the structural stress evolution tension function Jstress, combined with the disturbance anomaly indicator function L{Gopt>F1} for weighted calculation, a normalized comprehensive risk score index is output. Furthermore, based on the critical state of structural damage of the membrane material of seaside wind power equipment in salt spray testing, a risk interval threshold is set and the risk is divided into three levels, corresponding to process responses such as factory release, recommended re-inspection, and forced retirement. Through digital quantitative scoring and graded response mode, the present invention not only improves the efficiency and accuracy of defect processing, but also enhances the practical adaptability of the system in highly corrosive salt spray environments. It is particularly suitable for intelligent maintenance scenarios of large-scale, unmanned equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a schematic diagram of the steps of a film surface defect detection method based on machine vision of the present invention; Figure 2This is a flow chart of a film surface defect detection system based on machine vision according to the present invention; Figure 3 Schematic diagram of the arrangement of the film material detection light source and visual sensor of the present invention. DETAILED DESCRIPTION

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0021] Example 1: Please refer to Figure 1 and Figure 3 The present invention provides a method for detecting surface defects of film materials based on machine vision. To achieve the above purpose, the present invention is implemented through the following technical solutions: comprising the following steps: S1. Setting a visual sensor to collect image data of the outer sealing film of the signal control unit of the wind turbine equipment and transmitting the image data of the outer sealing film to the defect detection cloud server; S2. Preprocess the outer film sealing image data in the defect detection cloud server to obtain a standardized image set, and then perform feature extraction on the standardized image set to obtain a disturbance data set; S3. Calculate and output the light reflection disturbance anomaly index function Gpot based on the disturbance data set, conduct preliminary comparative evaluation, preliminarily judge the salt spray disturbance, and trigger the stress evolution trend inference mechanism; S4. After triggering the stress evolution trend inference mechanism, calculate and output the structural stress evolution tension function Jstress, and calculate and output the membrane material corrosion risk function Rtotal based on the structural stress evolution tension function Jstress; S5. Set the risk interval threshold, conduct a secondary comparative evaluation of the membrane material corrosion risk function Rtotal and the risk interval threshold, and classify the secondary comparative evaluation results and implement the corresponding response mechanism.

[0022] In this embodiment, the method, through the sequential implementation of steps S1 to S5, establishes an integrated detection process from image acquisition, standardized preprocessing, disturbance parameter extraction, risk index calculation, level determination, to intelligent response. Specifically, by combining a dual-angle light source with a visual sensor for image acquisition, the compatibility and imaging balance of different reflective characteristic regions (specular / diffuse reflection) of the film material are improved. Multi-frame fusion and filtering are used to effectively eliminate artifacts and dust particle interference during the image preprocessing stage, resulting in a standardized image set with uniform grayscale and prominent texture. Based on the standardized image set, a disturbance data set is extracted and the output light reflection disturbance anomaly index function Gpot is calculated to construct a "latent corrosion map" reflecting the early salt spray corrosion tendency. The structural stress evolution tension function Jstress is further calculated to accurately analyze the structural evolution trend of potential defects. Finally, the light reflection disturbance anomaly index function Gpot and the structural stress evolution tension function Jstress are combined to output the film material corrosion risk function Rtotal. The risk interval threshold is set based on the structural failure critical point, achieving multi-level determination of the film material status and a differentiated treatment strategy. The present invention significantly improves the sensitivity and predictive power of membrane defect detection in salt spray corrosion scenarios. Compared to traditional static detection techniques based on the naked eye, single-frame images, or contour differentials, the present invention can capture dynamic trends and conduct multi-dimensional information fusion analysis. Its beneficial effects are reflected not only in the high-confidence identification of early salt corrosion points, but also in the intelligent classification of early warning and response decisions for potential structural instability areas. This provides reliable intelligent detection assurance and maintenance reference for the long-term safe operation of wind turbines in highly corrosive coastal environments.

[0023] Example 2: Please refer to Figure 1 and Figure 3 ,Specifically: S1 includes S11 and S12; S11. At the outer sealing film of the communication and signal control unit of the seaside wind turbine equipment, a fixed dual-angle light source is simultaneously set to provide fill light for the outer sealing film, and a visual sensor is used to perform multi-frame sampling and shooting of the outer sealing film to obtain image data of the outer sealing film in real time; The fixed dual-angle light source includes a main light source and an auxiliary light source. The main light source is arranged at a 15° angle to emphasize specular reflection, while the auxiliary light source is arranged at a 45° angle to emphasize diffuse reflection and scattering information. At the same time, a 5600K white light LED light source is used in conjunction with a diffuse reflection background plate to eliminate external stray light. Multi-frame sampling shooting continuously captures 3 frames of outer film image data of each communication and signal control unit to eliminate artifacts and dust interference; S12. Through the communication module of the visual sensor, use the 5G wireless communication network to wirelessly integrate the visual sensor with the defect detection cloud server, and transmit the outer sealing film image data collected in real time to the defect detection cloud server.

[0024] In this embodiment, the method introduces a composite image acquisition mechanism of "fixed dual-angle light source and multi-frame sampling + high-speed wireless transmission" in the front-end link of film material defect detection. Specifically, a main and auxiliary dual light source system is arranged at the outer sealing film of the communication and signal control unit, wherein the main light source is arranged at 15° to highlight the mirror reflection characteristics, and the auxiliary light source is arranged at 45° to enhance the diffuse reflection and scattering information. In conjunction with the 5600K white light LED and the diffuse reflection background plate, the interference of ambient stray light is effectively eliminated, and the high-fidelity acquisition of the reflection characteristics of different surfaces is achieved; at the same time, the 3-frame multi-frame continuous sampling technology is adopted, and the image fusion within the time window is used to improve the ability to eliminate non-structural artifacts such as floating dust particles, water mist, and light spots, ensuring the stability and consistency of image quality. In addition, by integrating a 5G communication module on the visual sensor, a high-speed and low-latency wireless upload channel for image data is built, so that a real-time intercommunication mechanism is formed between the image acquisition module and the defect detection cloud server, providing sufficient bandwidth and data timeliness for subsequent image processing and defect analysis algorithms. This implementation method achieves the goal of high-fidelity, high-robustness, and high-efficiency original image acquisition, providing a solid foundation for high-precision detection of surface defects in film materials.

[0025] Example 3: Please refer to Figure 1 , specifically: S2 includes S21 and S22; S21. Receiving outer film sealing image data in real time in the defect detection cloud server, and preprocessing the outer film sealing image data to obtain a standardized image set; Preprocessing includes denoising, block processing, multi-frame fusion, and grayscale and contrast normalization; Denoising uses bilateral filtering, Gaussian filtering, and Laplace inversion to remove background low-frequency distribution while retaining edge information, and is used to remove high-frequency interference such as light spots, ambient interference light, and thermoelectric noise; Therefore, the block processing is performed by cutting the outer film image data into 64×64 blocks, and setting the lower left corner of the outer film image data to the pixel coordinate vertex (0,0). The horizontal and vertical axes of each block constitute the pixel coordinates (x, y), where x represents the horizontal pixel coordinate and y represents the vertical pixel coordinate. Grayscale and contrast normalization: By using histogram equalization, adaptive contrast optimization, and Min-Max grayscale normalization, the grayscale normalization of the block-processed outer film image data is performed to ensure that the brightness range of all images is consistent and adaptable to different materials, such as transparent film, matte film, etc. Multi-frame fusion simultaneously analyzes the mean and temporal variance of the pixels in the time window for three frames of outer film image data collected continuously from the same viewing angle. It also sets a floating artifact judgment threshold to remove outer film image data with values higher than the mean and temporal variance. This is used to filter out non-structural false defects such as floating dust particles, water mist, and interfering light spots. S22. Perform feature extraction on the standardized image set using an image feature extraction technique to obtain a perturbation dataset, and perform Z-score standardization on the obtained perturbation dataset, subtracting the mean and dividing the standard deviation to obtain a standard normal distribution of the parameters in the perturbation dataset, thereby eliminating the dimensionality effect of the parameters in the perturbation dataset; The perturbation data set includes the local reflectance difference △Ploc(x,y) of the pixel coordinate (x,y), the local stacking perturbation index S(x,y) of the pixel coordinate (x,y), the local grayscale variance Wvar(x,y) of the pixel coordinate (x,y), the local structure residual tensor field Tres(x,y) of the pixel coordinate (x,y), the local gradient standard deviation Jb(x,y) of the pixel coordinate (x,y) and the surface morphological curvature K(x,y) of the pixel coordinate (x,y); The local reflectivity difference △Ploc(x, y) at the pixel coordinate (x, y) is obtained by performing a normalized difference calculation on two grayscale outer film images taken with a fixed dual-angle light source. This value is used to reflect local light reflection anomalies caused by salt crystals, starting corrosion points, pits, etc., and is one of the core indicators for early judgment of salt corrosion. The local stacked perturbation index S(x, y) of the pixel coordinate (x, y) is obtained by using the OpenCV convolution window operation. A 5×5 domain window is taken for the center point of each pixel coordinate (x, y) in the standardized image set. The domain window is divided into upper and lower parts and left and right parts along the symmetry axis. The sum of squared recovery differences between the upper and lower parts and the left and right parts is calculated to filter out homogeneous high-reflective areas such as material edges, indentations, and light source mapping, and focus on structural asymmetric defects. The local grayscale variance Wvar(x, y) of the pixel coordinate (x, y) is calculated by NumPy and OpenCV sliding window variance, and the grayscale mean and standard deviation of the domain window are calculated; The local structure residual tensor field Tres(x, y) of the pixel coordinates (x, y) is extracted by using OpenCV and SciPy to normalize the tensor of each image in the image set, and then the divergence is extracted using the Sobel algorithm to obtain whether the area is releasing or accumulating structural deformation tension, which is the core trigger signal for judging whether it is "evolving"; The local gradient standard deviation Jb(x, y) of the pixel coordinate (x, y) is obtained by calculating the grayscale gradient map using the Sobel operator. The gradient standard deviation is then calculated for each central pixel within the domain window using OpenCV. This is used in conjunction with the structural residual tensor to improve the ability to identify "microcrack edge instability" areas. The surface morphological curvature K(x, y) of the pixel coordinate (x, y) is obtained by using the Hessian matrix to analyze the curvature in the horizontal pixel coordinate x direction and the vertical pixel coordinate y direction, and is calculated and output based on Gaussian curvature. It is used to identify the micro-undulations and mutation locations of the film material, which helps to find the geometric trap points where salt spray accumulates in the material microstructure.

[0026] In this embodiment, the method establishes a complete standardized mechanism from image preprocessing to disturbance feature extraction in the defect detection data processing link. First, after receiving the outer film sealing image data at the defect detection cloud server, multiple image preprocessing strategies are adopted, including a multimodal denoising technology combining bilateral filtering, Gaussian filtering and Laplace inversion to effectively remove thermoelectric noise, ambient stray light and light spot artifacts, while retaining edge and detail information to ensure the clarity and structural fidelity of image quality. Subsequently, by standardizing and cutting the image into 64×64 pixel blocks and introducing histogram equalization, adaptive contrast enhancement and Min-Max normalization operations, the image is unified in grayscale and contrast, improving the algorithm's adaptability and robustness to the reflective properties of different film surfaces, such as transparent films and matte films. Multi-frame fusion further enhances image stability, and non-structural pseudo-defects such as floating dust and water mist are identified and eliminated through time domain mean and variance analysis, greatly improving the retention rate of effective feature information. On this basis, a perturbation dataset was constructed using image feature extraction techniques, extracting several key image physical features, including the local reflectivity difference ΔPloc, the perturbation symmetry index S, the local grayscale variance Wvar, the structural residual tensor Tres, the local gradient standard deviation Jb, and the surface morphological curvature K. These feature parameters were normalized using the Z-score to eliminate dimensionality effects, making subsequent algorithm processing more stable, comparable, and statistically robust. Each feature has clear defect indicative significance, such as ΔPloc, which directly reflects potential salt corrosion initiation points; Tres and Jb, coupled to identify structural edge evolution trends; and K, which helps locate geometric trap zones.

[0027] Example 4: Please refer to Figure 1 ,Specifically: S3 includes S31 and S32; S31. Constructing a light reflection disturbance risk algorithm model, extracting a disturbance data set and inputting it into the light reflection disturbance risk algorithm model, calculating and outputting a light reflection disturbance anomaly index function Gpot, and measuring the optical disturbance index present in the collected outer film image data; The light reflection disturbance anomaly index function Gpot is calculated and output by the following light reflection disturbance risk algorithm model; ; Where Gpot(x, y) represents the light reflection disturbance anomaly exponential function of the pixel coordinate (x, y), log represents the logarithmic function, represents the anti-divergence constant, which is set to 1×10 -5 To avoid denominators being zero, the logarithmic function log is used to calculate anomalies; Indicates suppression of texture false positives. The greater the texture disturbance, the stronger the suppression, preventing the texture boundary from being misjudged as a defect. The molecular term is a joint sensitivity factor of light reflection and structural disturbance, taking both brightness and structural abnormalities into account. It is the key to measuring whether the pixel area may have "optical abnormality defects, such as the starting point of salt corrosion." It can also determine whether a pixel is abnormal due to reflectivity differences or optical changes; and whether it is located in an asymmetric structural disturbance area, which may be a damage or corrosion point. The physical significance of the formula lies in using light reflection differences and structural asymmetry to identify areas where salt spray corrosion may occur, while adjusting the recognition sensitivity by adjusting the local texture complexity to construct a potential map of corrosion points.

[0028] S32. After obtaining the light reflection disturbance anomaly index function Gpot(x, y) of the pixel coordinates (x, y) of each block, perform a preliminary comparative evaluation. The preliminary comparative evaluation is performed by sampling a set of outer film image data samples containing known defects, and calculating and outputting the light reflection disturbance anomaly index function Gpot(x, y) of the pixel coordinates (x, y). The light reflection disturbance anomaly index function Gpot(x, y) of the minimum pixel coordinates (x, y) is selected and set as the disturbance threshold F1. The disturbance threshold F1 is compared with the light reflection disturbance anomaly index function Gpot(x, y) of the pixel coordinates (x, y) of all blocks to preliminarily determine the characteristics of salt spray erosion and trigger the stress evolution trend inference mechanism. The specific evaluation content is as follows; When the light reflection disturbance anomaly index function Gpot(x, y) of the pixel coordinate (x, y) is greater than the disturbance threshold F1, it indicates disturbance anomaly. It is preliminarily judged that salt spray disturbance exists. At this time, the proportion of disturbance anomalies in all the current outer film image data blocks is analyzed. If the proportion of disturbance anomalies exceeds 30%, the stress evolution trend inference mechanism is triggered; If the proportion of disturbance anomalies is less than or equal to 30%, the detection frequency will be increased by 50%; When the light reflection disturbance anomaly index function Gpot(x, y) of the pixel coordinate (x, y) is ≤ the disturbance threshold F1, it means that the disturbance is normal and does not belong to the corrosion risk area. At this time, the next block is traversed.

[0029] In this embodiment, this method systematically quantitatively identifies potential optical anomalies on the film surface and provides preliminary risk assessment by constructing a light reflection perturbation risk algorithm model and using the previously extracted perturbation dataset as input. First, using the light reflection perturbation anomaly index function (Gpot), the combined sensitivity factor of local reflectivity differences and structural perturbations, combined with nonlinear suppression of texture complexity, achieves highly sensitive extraction of optical anomalies on the film surface caused by factors such as salt crystals, micro-pits, and initial corrosion pits. This algorithm model avoids mathematical divergence by introducing micro-constants and amplifies abnormal signals using a logarithmic scale, effectively improving the ability to identify early areas of light reflection asymmetry and providing a solid foundation for the construction of corrosion pit potential maps. Subsequently, in S32, the light reflection perturbation anomaly index function (Gpot) is calculated for a set of training sample images containing real defects, and a perturbation threshold (F1) is extracted. The light reflection perturbation anomaly index function (Gpot) value for each pixel in the real-time image is then classified and assessed. When the light reflection perturbation anomaly index function (Gpot) exceeds the perturbation threshold (F1), the region is marked as a perturbation anomaly, and the anomaly percentage is calculated. When the anomaly ratio exceeds 30%, the subsequent stress evolution trend inference mechanism is automatically triggered to realize the linkage judgment from optical signals to structural evolution; if the ratio is low, the system dynamically increases the detection frequency to form a detection strategy based on adaptive adjustment of the disturbance degree.

[0030] Example 5: Please refer to Figure 1 , specifically: S4 includes S41 and S42; S41. After preliminary comparative evaluation of the trigger stress evolution trend inference mechanism, the disturbance data set is extracted and the structural stress evolution tension function Jstress is calculated and output to analyze the evolution of defects caused by salt spray erosion. The structural stress evolution tension function Jstress is calculated and output by the following algorithm formula; ; Where Jstress(x, y) represents the structural stress evolution tension function of the pixel coordinate (x, y), represents the divergence of the tensor field; Indicates the amount of local stress release. The divergence in the tensor field is a "source / sink" indicator. In the image, it manifests as a sudden expansion or concentration in the texture direction, such as the beginning of the expansion of the scratch, the formation of fine cracks around the erosion point, and edge instability. The larger the value, the more imbalanced the local structure in the area is, which is a core indicator of whether the potential crack initiation point is tension release or contraction; It represents the boundary activity and curvature evolution factor. The product of the two is a nonlinear enhancement. When "severe boundary disturbance" and "obvious geometric deformation" are present at the same time, the result will be significantly amplified, ordinary texture fluctuations will be filtered out, and only the budding characteristics of structural defects will be captured.

[0031] S42, based on the obtained structural stress evolution tension function Jstress and light reflection disturbance anomaly index function Gpot, a comprehensive calculation is performed to output a film material corrosion risk function Rtotal, and a comprehensive analysis is performed on the comprehensive erosion and evolution trend risk in each outer film image data; The membrane corrosion risk function Rtotal is calculated and output by the following algorithm formula; ; Where A represents the total number of all blocks assessed as disturbance anomalies, L{Gopt>F1} represents the indicator function, a1 and a2 represent the preset weight values of the light reflection disturbance anomaly index function Gpot and the structural stress evolution tension function Jstress, respectively. The specific values are set by the user, and a1+a2=1; The calculation logic calculates the first term for each pixel coordinate (x, y) , to determine whether If so, multiply by The normalized film corrosion risk function Rtotal is obtained by summing up all pixel coordinates (x, y) and dividing it by the total number A of all blocks assessed as disturbance anomalies; It represents a weighted average of the risk scores of the pixel coordinates (x, y) of all blocks to obtain a unified global score, which is used for risk assessment of the entire outer film image data, regional risk ranking, and level classification; Indicates whether each pixel has asymmetric disturbance, grayscale texture fluctuation, and light reflectivity difference in optical angle, that is, whether it is a potential salt corrosion initiation point. The weight value a1 of the light reflectance disturbance anomaly index function is used to control the weight of such optical features in the overall risk assessment; A dynamic structural assessment mechanism that only performs in-depth analysis on truly suspicious points, avoiding redundant calculations while focusing on structural defects. Represents a Boolean judgment function. If the light reflection disturbance anomaly index function Gpot(x, y) of the pixel coordinate (x, y) exceeds the disturbance threshold F1, it is 1, otherwise it is 0; It is used to judge whether these high-risk points have a structural evolution trend; The weight value a2 of the structural stress evolution tension function is used to control the contribution of the structural evolution factor in the total score; The overall physical meaning of the formula is to allow early characteristics to control warning sensitivity and later trends to control severity judgment, ultimately integrating them into a unified, quantifiable risk score. Therefore, the physical meaning can be expressed as: a normalized regional risk value, which comprehensively represents the possibility of corrosion in the current state of the membrane material, the "starting point", and the trend of further deterioration, facilitating the decision on whether to repair, issue a warning, or conduct a review.

[0032] In this embodiment, this method further deepens the analysis and judgment of salt spray corrosion evolution trends in membrane materials by constructing a structural stress evolution inference mechanism. In S41, based on the perturbed dataset, the structural stress evolution tension function Jstress is calculated and output. This function integrates the divergence term of the image texture residual tensor field with local geometric characteristic factors, grayscale gradient standard deviation, and surface curvature. Through nonlinear enhancement, it effectively highlights areas of structural stress imbalance. This method can identify potential crack initiation areas, boundary microcracks, and material geometric deformation "trap sites" caused by initial salt spray corrosion, demonstrating highly sensitive structural evolution detection capabilities. Furthermore, in S42, a membrane corrosion risk function Rtotal is constructed, implementing a global quantitative risk assessment mechanism that weightedly integrates early optical anomalies with later structural stress evolution. Rtotal is calculated by applying the structural assessment function to each high-risk optical point and normalizing it using the total number of abnormal blocks A as a reference to output a unified normalized corrosion risk value. This value exhibits good sensitivity and discrimination, and can be used for image-wide corrosion trend grading and risk ranking. The introduction of weight values allows for flexible adjustment of warning sensitivity and trend judgment intensity based on material type or usage environment.

[0033] Example 6: Please refer to Figure 1 ,Specifically: S5 includes S51 and S52; S51, reversely setting the risk interval threshold according to the critical state of structural damage of the membrane material in the salt spray test, the risk interval threshold including the risk upper limit threshold Q1 and the risk lower limit threshold Q2; Among them, the upper risk threshold Q1 is the critical value between erosion and defects; the lower risk threshold Q2 is the critical value between normal and erosion; The real-time obtained membrane corrosion risk function Rtotal is compared and evaluated with the risk interval threshold to determine the overall membrane corrosion risk of the currently acquired outer film image data. The evaluation is then graded based on the secondary comparison and evaluation results. The specific evaluation contents are as follows; When the membrane material corrosion risk function Rtotal is less than the lower risk threshold Q2, the outer sealing membrane of the communication and signal control unit of the current seaside wind power equipment is classified as level one risk; When the lower risk threshold Q2 ≤ membrane corrosion risk function Rtotal < upper risk threshold Q1, the outer sealing membrane of the communication and signal control unit of the current seaside wind power equipment is classified as a secondary risk; When the membrane material corrosion risk function Rtotal ≥ the risk upper limit threshold Q1, the outer sealing membrane of the communication and signal control unit of the current seaside wind power equipment is divided into the third level of risk.

[0034] S52. Based on the secondary comparative evaluation, the corresponding response measures are implemented based on the classification of the outer sealing film of the communication and signal control unit of the current seaside wind power equipment. The specific response measures are as follows; When classified as level one risk, it means that the outer sealing film of the communication and signal control unit of the current seaside wind power equipment is blocked by salt fog. At this time, no management is required and it should be cleaned as planned; When it is classified as a Level 2 risk, it means that the outer sealing membrane of the communication and signal control unit of the current offshore wind turbine equipment is corroded by salt spray, but there are no structural defects. At this time, an early warning prompt is generated and the outer sealing membrane of the communication and signal control unit of the offshore wind turbine equipment is immediately cleaned of salt spray. When it is classified as the third level of risk, it means that the outer sealing film of the communication and signal control unit of the current seaside wind power equipment is corroded by salt spray, resulting in defects in the membrane material structure. At this time, the operation of the current seaside wind power equipment is stopped, and an early warning is generated to replace the outer sealing film of the communication and signal control unit.

[0035] In this embodiment, this method utilizes a graded assessment and response mechanism based on the membrane corrosion risk function Rtotal, achieving a closed-loop process from intelligent identification to orderly control. In S51, the system sets an upper risk threshold (Q1) (defect threshold) and a lower risk threshold (Q2) (erosion threshold) based on the membrane's critical structural damage status during salt spray testing. By performing a secondary comparison between the membrane corrosion risk function Rtotal and these two thresholds, a precise risk classification is achieved. Specifically, the membrane status is divided into three levels: level 1, level 2, and level 3. The corresponding risk level and treatment requirements for each level are clearly defined, ensuring the practicality and operability of the judgment results. In S52, a corresponding response strategy is automatically matched to each risk level. For level 1 risk, the equipment is allowed to continue operating, with only scheduled cleaning. For level 2 risk, a timely warning is generated and manual cleaning is initiated. For level 3 risk, operation is immediately stopped and replacement is triggered to prevent serious consequences such as membrane perforation and short circuits during operation. This strategy enhances proactive intervention capabilities and avoids further damage caused by delayed response.

[0036] Example 7: Please refer to Figure 1 and Figure 2 , a film surface defect detection system based on machine vision, including a visual acquisition module, a film data extraction module, a salt spray disturbance analysis module, a disturbance evolution trend analysis module and a defect response module; The visual acquisition module collects the image data of the outer sealing film of the signal control unit of the wind turbine equipment by setting up a visual sensor, and transmits the image data of the outer sealing film to the defect detection cloud server; The film material data extraction module pre-processes the outer film sealing image data in the defect detection cloud server to obtain a standardized image set, and then performs feature extraction on the standardized image set to obtain a disturbance data set; The salt fog disturbance analysis module calculates and outputs the light reflection disturbance anomaly index function Gpot based on the disturbance data set, performs preliminary comparative evaluation, preliminarily determines the salt fog disturbance, and triggers the stress evolution trend inference mechanism; The disturbance evolution trend analysis module calculates and outputs the structural stress evolution tension function Jstress after triggering the stress evolution trend inference mechanism. Based on the structural stress evolution tension function Jstress, it calculates and outputs the membrane material corrosion risk function Rtotal. The defect response module sets the risk interval threshold, conducts a secondary comparative evaluation of the membrane material corrosion risk function Rtotal and the risk interval threshold, and classifies the secondary comparative evaluation results and executes the corresponding response mechanism.

[0037] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

Claims

1. A method for detecting surface defects of film materials based on machine vision, characterized by: The following steps are involved: S1. Setting a visual sensor to collect image data of the outer sealing film of the signal control unit of the wind turbine equipment and transmitting the image data of the outer sealing film to the defect detection cloud server; S2. Preprocess the outer film sealing image data in the defect detection cloud server to obtain a standardized image set, and then perform feature extraction on the standardized image set to obtain a disturbance data set; S3. Calculate and output the light reflection disturbance anomaly index function Gpot based on the disturbance data set, conduct preliminary comparative evaluation, preliminarily judge the salt spray disturbance, and trigger the stress evolution trend inference mechanism; S4. After triggering the stress evolution trend inference mechanism, calculate and output the structural stress evolution tension function Jstress, and calculate and output the membrane material corrosion risk function Rtotal based on the structural stress evolution tension function Jstress; S5. Set the risk interval threshold, conduct a secondary comparative evaluation of the membrane material corrosion risk function Rtotal and the risk interval threshold, and classify the secondary comparative evaluation results and implement the corresponding response mechanism.

2. The method for detecting film surface defects based on machine vision according to claim 1, characterized in that: Said S1 includes S11 and S12; S11. At the outer sealing film of the communication and signal control unit of the seaside wind turbine equipment, a fixed dual-angle light source is simultaneously set to provide fill light for the outer sealing film, and a visual sensor is used to perform multi-frame sampling and shooting of the outer sealing film to obtain image data of the outer sealing film in real time; The fixed dual-angle light source includes a main light source and an auxiliary light source; the main light source is arranged at a 15° angle to emphasize specular reflection, and the auxiliary light source is arranged at a 45° angle to emphasize diffuse reflection and scattering information. At the same time, a 5600K white light LED light source is used in conjunction with a diffuse reflection background plate to eliminate external stray light; The multi-frame sampling shooting is performed by continuously shooting 3 frames of outer sealing film image data at the outer sealing film of each communication and signal control unit; S12. Through the communication module of the visual sensor, use the 5G wireless communication network to wirelessly integrate the visual sensor with the defect detection cloud server, and transmit the outer sealing film image data collected in real time to the defect detection cloud server.

3. The method for detecting film surface defects based on machine vision according to claim 2, characterized in that: Said S2 includes S21 and S22; S21. Receiving outer film sealing image data in real time in the defect detection cloud server, and preprocessing the outer film sealing image data to obtain a standardized image set; The preprocessing includes denoising, block processing, multi-frame fusion and grayscale and contrast normalization; The denoising is performed by bilateral filtering, Gaussian filtering and Laplace inversion to remove background low-frequency distribution while retaining edge information; Therefore, the block processing is performed by cutting the outer film image data into 64×64 blocks, and setting the lower left corner of the outer film image data to the pixel coordinate vertex (0,0). The horizontal and vertical axes of each block constitute the pixel coordinates (x, y), where x represents the horizontal pixel coordinate and y represents the vertical pixel coordinate. The grayscale and contrast normalization is performed on the outer sealing film image data processed in blocks by using histogram equalization, adaptive contrast optimization and Min-Max grayscale normalization; The multi-frame fusion simultaneously performs mean and temporal variance analysis on the pixels of the three frames of outer film image data collected continuously from the same viewing angle within the time window, and sets a floating artifact judgment threshold to eliminate the outer film image data with a value higher than the mean and temporal variance; S22. Perform feature extraction on the standardized image set using an image feature extraction technique to obtain a perturbation dataset, and perform Z-score standardization on the obtained perturbation dataset, subtracting the mean and dividing the standard deviation to obtain a standard normal distribution of the parameters in the perturbation dataset, thereby eliminating the dimensionality effect of the parameters in the perturbation dataset; The perturbation data set includes the local reflectivity difference △Ploc(x,y) of the pixel coordinate (x, y), the local stacking perturbation index S(x, y) of the pixel coordinate (x, y), the local grayscale variance Wvar(x, y) of the pixel coordinate (x, y), the local structure residual tensor field Tres(x, y) of the pixel coordinate (x, y), the local gradient standard deviation Jb(x, y) of the pixel coordinate (x, y) and the surface morphological curvature K(x, y) of the pixel coordinate (x, y).

4. The method for detecting film surface defects based on machine vision according to claim 3, characterized in that: Said S3 includes S31 and S32; S31. Constructing a light reflection disturbance risk algorithm model, extracting a disturbance data set and inputting it into the light reflection disturbance risk algorithm model, calculating and outputting a light reflection disturbance anomaly index function Gpot, and measuring the optical disturbance index present in the collected outer film image data; The light reflection disturbance anomaly index function Gpot is calculated and output by the following light reflection disturbance risk algorithm model; ； Where Gpot(x, y) represents the light reflection disturbance anomaly exponential function of the pixel coordinate (x, y), log represents the logarithmic function, represents the anti-divergence constant, which is set to 1×10 -5 Avoid denominators of 0.

5. The method for detecting film surface defects based on machine vision according to claim 4, characterized in that: S32. After obtaining the light reflection disturbance anomaly index function Gpot(x, y) of the pixel coordinates (x, y) of each block, perform a preliminary comparative evaluation. The preliminary comparative evaluation is performed by sampling a set of outer film image data samples containing known defects, and calculating and outputting the light reflection disturbance anomaly index function Gpot(x, y) of the pixel coordinates (x, y). The light reflection disturbance anomaly index function Gpot(x, y) of the minimum pixel coordinates (x, y) is selected and set as the disturbance threshold F1. The disturbance threshold F1 is compared with the light reflection disturbance anomaly index function Gpot(x, y) of the pixel coordinates (x, y) of all blocks to preliminarily determine the characteristics of salt spray erosion and trigger the stress evolution trend inference mechanism. The specific evaluation content is as follows; When the light reflection disturbance anomaly index function Gpot(x, y) of the pixel coordinate (x, y) is greater than the disturbance threshold F1, it indicates disturbance anomaly. It is preliminarily judged that salt spray disturbance exists. At this time, the proportion of disturbance anomalies in all the current outer film image data blocks is analyzed. If the proportion of disturbance anomalies exceeds 30%, the stress evolution trend inference mechanism is triggered; If the proportion of disturbance anomalies is less than or equal to 30%, the detection frequency will be increased by 50%; When the light reflection disturbance anomaly index function Gpot(x, y) of the pixel coordinate (x, y) is ≤ the disturbance threshold F1, it means that the disturbance is normal and does not belong to the corrosion risk area. At this time, the next block is traversed.

6. The method for detecting film surface defects based on machine vision according to claim 5, characterized in that: Said S4 includes S41 and S42; S41. After preliminary comparative evaluation of the trigger stress evolution trend inference mechanism, the disturbance data set is extracted and the structural stress evolution tension function Jstress is calculated and output to analyze the evolution of defects caused by salt spray erosion. The structural stress evolution tension function Jstress is calculated and outputted by the following algorithm formula; ； Where Jstress(x, y) represents the structural stress evolution tension function of the pixel coordinate (x, y), Represents the divergence of the tensor field.

7. The method for detecting film surface defects based on machine vision according to claim 6, characterized in that: S42, based on the obtained structural stress evolution tension function Jstress and light reflection disturbance anomaly index function Gpot, a comprehensive calculation is performed to output a film material corrosion risk function Rtotal, and a comprehensive analysis is performed on the comprehensive erosion and evolution trend risk in each outer film image data; The membrane material corrosion risk function Rtotal is calculated and outputted by the following algorithm formula: ； Where A represents the total number of all blocks assessed as disturbance anomalies, L{Gopt>F1} represents the indicator function, a1 and a2 represent the preset weight values of the light reflection disturbance anomaly index function Gpot and the structural stress evolution tension function Jstress, respectively. The specific values are set by the user, and a1+a2=1.

8. The method for detecting film surface defects based on machine vision according to claim 7, characterized in that: Said S5 includes S51 and S52; S51, reversely setting a risk interval threshold based on the critical state of structural damage of the membrane material in the salt spray test, wherein the risk interval threshold includes an upper risk threshold Q1 and a lower risk threshold Q2; Among them, the upper risk threshold Q1 is the critical value between erosion and defects; the lower risk threshold Q2 is the critical value between normal and erosion; The real-time obtained membrane corrosion risk function Rtotal is compared and evaluated with the risk interval threshold to determine the overall membrane corrosion risk of the currently acquired outer film image data. The evaluation is then graded based on the secondary comparison and evaluation results. The specific evaluation contents are as follows; When the membrane material corrosion risk function Rtotal is less than the lower risk threshold Q2, the outer sealing membrane of the communication and signal control unit of the current seaside wind power equipment is classified as level one risk; When the lower risk threshold Q2 ≤ membrane corrosion risk function Rtotal < upper risk threshold Q1, the outer sealing membrane of the communication and signal control unit of the current seaside wind power equipment is classified as a secondary risk; When the membrane material corrosion risk function Rtotal ≥ the risk upper limit threshold Q1, the outer sealing membrane of the communication and signal control unit of the current seaside wind power equipment is divided into the third level of risk.

9. The method for detecting film surface defects based on machine vision according to claim 8, characterized in that: S52. Based on the secondary comparative evaluation, the corresponding response measures are implemented based on the classification of the outer sealing film of the communication and signal control unit of the current seaside wind power equipment. The specific response measures are as follows; When classified as level one risk, it means that the outer sealing film of the communication and signal control unit of the current seaside wind power equipment is blocked by salt fog. At this time, no management is required and it should be cleaned as planned; When it is classified as a Level 2 risk, it means that the outer sealing membrane of the communication and signal control unit of the current offshore wind turbine equipment is corroded by salt spray, but there are no structural defects. At this time, an early warning prompt is generated, and the outer sealing membrane of the communication and signal control unit of the offshore wind turbine equipment is immediately cleaned of salt spray. When it is classified as the third level of risk, it means that the outer sealing film of the communication and signal control unit of the current seaside wind power equipment is corroded by salt spray, resulting in defects in the membrane material structure. At this time, the operation of the current seaside wind power equipment is stopped, and an early warning is generated to replace the outer sealing film of the communication and signal control unit.

10. A film surface defect detection system based on machine vision, applied to the film surface defect detection method based on machine vision according to any one of claims 1 to 9, characterized in that: It includes visual acquisition module, film material data extraction module, salt spray disturbance analysis module, disturbance evolution trend analysis module and defect response module; The visual acquisition module collects image data of the outer sealing film of the signal control unit of the wind power equipment by setting a visual sensor, and transmits the image data of the outer sealing film to the defect detection cloud server; The film material data extraction module pre-processes the outer film sealing image data in the defect detection cloud server to obtain a standardized image set, and then extracts features from the standardized image set to obtain a disturbance data set; The salt fog disturbance analysis module calculates and outputs the light reflection disturbance anomaly index function Gpot based on the disturbance data set, performs preliminary comparative evaluation, preliminarily determines the salt fog disturbance, and triggers the stress evolution trend inference mechanism; The disturbance evolution trend analysis module calculates and outputs the structural stress evolution tension function Jstress after triggering the stress evolution trend inference mechanism, and calculates and outputs the membrane material corrosion risk function Rtotal based on the structural stress evolution tension function Jstress; The defect response module sets a risk interval threshold, performs a secondary comparative evaluation on the film material corrosion risk function Rtotal and the risk interval threshold, and classifies the secondary comparative evaluation results and executes a corresponding response mechanism.

Images