Intelligent color difference analysis and detection method for paint

Abstract

The present application relates to the field of quality detection technology in material manufacturing and coating process, and more particularly to a kind of intelligent color difference analysis detection method for paint, comprising: step S1: multispectral dynamic compensation image acquisition;Step S2: texture-color coupling separation;Step S3: adaptive color difference modeling;Step S4: process correlation visualization, color difference data is mapped as scalable heat map, grid density is dynamically adjusted according to detection accuracy;The geometric characteristic parameters of abnormal area are analyzed, the process defect mode library is matched to generate optimization suggestions, and the heat map is overlaid with visualized identification. Through real-time data acquisition, texture interference elimination, model adaptive optimization and process defect analysis, the overall detection accuracy and efficiency are improved, and the intelligent adaptive characteristics are also possessed, which can adapt to different production environments and substrate types, thereby optimizing the paint production process and ensuring the consistency of coating quality.

Description

Technical Field

[0001] The present invention relates to the technical field of quality inspection in material manufacturing and coating processes, and in particular to an intelligent color difference analysis and detection method for coatings. Background Art

[0002] In the field of quality inspection technology in paint manufacturing and coating processes, color difference detection is a core link in quality control, directly affecting product appearance consistency and market competitiveness. The current mainstream color difference analysis technology mainly relies on fixed-point measurement with a colorimeter or image analysis methods based on standard colorimetric models. However, in actual application, there are the following significant defects:

[0003] Traditional colorimeters use a fixed light source incident angle and intensity for measurement, making them unable to adapt to the varying reflective properties of different coatings. For example, highly reflective coatings like metallic paint produce strong specular reflections, causing the measured value to deviate from the actual visual effect. Meanwhile, diffusely reflective surfaces like suede paint can experience color fluctuations due to interference from ambient stray light. While existing technologies attempt to reduce the impact of ambient light through the use of light shields, they are unable to dynamically compensate for differences in reflectivity across different areas. This results in poor repeatability when measuring the same coating surface (the ΔE fluctuation range often exceeds ±2.0), severely restricting the reliability of test results.

[0004] Texture features on coating surfaces (such as the natural grain of wood and the machined grain of metal) can optically couple with color information, leading to discrepancies between the results of color difference detection methods based on standard color difference models like CIELAB and the subjective evaluation of the human eye. Studies have shown that the accuracy of traditional color difference detection methods for complex coating surfaces is less than 82%. In high-end applications such as automotive interiors and furniture, the misjudgment rate is as high as 18%-25%, forcing companies to invest in additional manual re-inspection costs. Summary of the Invention

[0005] Based on the above purpose, the present invention provides an intelligent color difference analysis and detection method for coatings, comprising the following steps:

[0006] Step S1: Multispectral dynamic compensation image acquisition: In a closed inspection chamber, a ring array of adjustable light sources is used to project multi-band light onto the coating surface. The light source intensity in each area is dynamically adjusted based on the reflectivity reference value obtained in the pre-scan. A spectroscopic imaging device is used to separate the reflected light into the human eye-sensitive band and the texture enhancement band, and multi-channel image data is captured synchronously to eliminate registration errors caused by mechanical switching. The reflectivity reference value is dynamically calculated based on the ratio of the light intensity of diffuse reflection to specular reflection.

[0007] Step S2: texture-color coupling separation, extracting substrate-related texture features from the texture-enhanced band image, matching filter parameters based on the pre-stored substrate type library, performing multi-scale convolution operation on the texture-enhanced band image through the filter to extract the frequency domain response matrix containing the substrate inherent texture features; tensor fusion of the frequency domain response matrix containing the substrate inherent texture features and the RGB color channel is performed to establish a modulation model of texture to color channel, and an iterative optimization algorithm is used to eliminate texture interference to generate a decoupled pure color layer; the iterative optimization takes the similarity of adjacent iteration chroma histograms as the termination condition;

[0008] Step S3: adaptive color difference modeling, matching the pre-trained substrate classification model according to the texture feature frequency energy distribution, and loading the initial parameters; adding the artificial review samples in real-time detection to the training set, and dynamically updating the model weight using the difference-driven online learning algorithm to output the color difference grade and spatial distribution data;

[0009] Step S4: process-related visualization, mapping the color difference data into a scalable heat map, and dynamically adjusting the grid density according to the detection accuracy; analyzing the geometric feature parameters of the abnormal area, matching the process defect mode library to generate optimization suggestions, and superimposing the visualized identification on the heat map.

[0010] Preferably, the specific implementation method of dynamically adjusting the light source intensity in step S1 includes:

[0011] In the pre-scanning stage, the light source at a preset angle with the normal of the coating surface irradiates the to-be-detected area, and the spectrophotometer measures the maximum light intensity value in the specular reflection direction and the average light intensity value in the diffuse reflection area respectively, and calculates the ratio of the two as the reflectivity reference value; in real-time detection, the current detection area is scanned in blocks, the deviation percentage of the reflected light intensity of each sub-area from the reference value is calculated, and when the deviation percentage exceeds a preset threshold, the light source controller generates adjustment instructions according to the deviation direction and amplitude: for the area with reflectivity higher than the reference value, the light source intensity at the corresponding position is reduced by a certain proportion; for the area with reflectivity lower than the reference value, the light source intensity is compensated by a certain proportion, until the deviation of the real-time reflected light intensity from the reference value is controlled within the allowable range.

[0012] Preferably, the weighted correction method of the reflectivity reference value includes:

[0013] The reference value is dynamically corrected according to the type of the coating substrate, the correction coefficient of the metal substrate is calculated by the correlation between the surface roughness and the intensity attenuation rate of the specular reflection light, and the correction coefficient of the non-metal substrate is determined by the relationship between the texture anisotropy index and the uniformity of the diffuse reflection light intensity; the roughness is quantified by the gray level gradient variance of the near-infrared image, and the anisotropy index is calculated by the standard deviation of the texture direction distribution histogram; the corrected reference value is used for adjusting the light source intensity in the real-time detection stage, so as to ensure the consistency of the reflectivity measurement of different substrates.

[0014] Preferably, the parameter matching method of the filter in step S2 comprises:

[0015] The typical texture feature data of metal, plastic and wood are stored in the pre-stored substrate feature library, and the data is obtained by the following method: multi-angle near-infrared imaging is performed on the standard sample of each substrate, and the texture spatial frequency spectrum and the direction distribution histogram are extracted; the determination condition for matching the high-frequency filter of the metal substrate is that the energy proportion of the spatial frequency spectrum in the preset high-frequency interval exceeds a first threshold value, the determination condition for matching the medium-frequency filter of the plastic substrate is that the number of main peaks of the direction distribution histogram is less than a second threshold value, and the determination condition for matching the multi-direction filter of the wood substrate is that the direction distribution standard deviation is greater than a third threshold value; during real-time detection, the filter parameter combination is automatically selected according to the frequency domain features of the current near-infrared image.

[0016] Preferably, the implementation process of the iterative optimization algorithm in step S2 comprises:

[0017] The texture coupling coefficient matrix is initialized, the texture feature map is convolved with the color channel image to generate an initial texture modulation model; the base layer is iteratively optimized by the least square method to minimize the mean square error of the texture modulation residual error; in each iteration, the cosine similarity of the chroma histogram between the current base layer and the previous iteration result is calculated, and the optimization is terminated when the similarity changes by less than a convergence threshold value for three consecutive iterations; the convergence threshold value is dynamically adjusted according to the color difference tolerance level of the coating, and the higher the color difference requirement, the higher the threshold value.

[0018] Preferably, the detailed process of matching the pre-trained substrate classification model in step S3 comprises:

[0019] The near-infrared image is subjected to fast Fourier transform to obtain a frequency energy distribution graph, and the energy proportion in the metal substrate characteristic frequency band and the non-metal substrate characteristic frequency band is calculated; the energy proportion is input into the pre-trained substrate classification model, and the classification model learns the frequency domain feature boundary conditions of different substrates through historical sample data; the substrate type with the highest matching degree score is output, and the corresponding pre-trained color difference model parameters are loaded; when the matching degree score is lower than a confidence threshold value, an artificial review process is triggered and new substrate feature data is recorded.

[0020] Preferably, the weight update strategy of the online learning algorithm in the step S3 comprises:

[0021] The difference between the color difference level of the new sample and the prediction result of the model is quantified as a loss function, and the difference is determined by calculating the Euclidean distance of the two in hue, saturation and lightness dimensions;The gradient direction is calculated according to the partial derivative of the loss function to the weight of the full connection layer of the model, and the weight update step is adjusted by using the momentum acceleration method;When the loss function of the continuous multiple new samples decreases at a rate exceeding the historical average level, the learning rate is automatically expanded to accelerate the model convergence.

[0022] Preferably, the heat map generation method in the step S4 comprises:

[0023] According to the spatial resolution of the high-speed image sensor, the actual physical size corresponding to a single pixel is calculated, the division density of the grid unit is determined, so that the length of each unit is equal to an integer multiple of the preset detection accuracy;The color difference of all pixel points in each grid unit is statistically analyzed, and the average value and standard deviation of the ΔE value are calculated;The discrete grid data is converted into continuous color difference distribution heat map by using bicubic interpolation algorithm, and different ΔE value intervals are reflected by using color gradient mapping.

[0024] Preferably, the construction method of the process defect mode library in the step S4 comprises:

[0025] The color difference distribution samples caused by abnormal spraying pressure, spraying gun moving speed deviation and improper paint viscosity in historical production are collected, and the following geometric features are extracted for each sample: area ratio of abnormal area, center of gravity position offset, edge sharpness index;The feature data is divided into three modes of ring distribution, linear distribution and scattered point distribution by clustering algorithm;For each mode, a process parameter adjustment rule library is established, the ring distribution is associated with the spraying pressure reduction ratio, the linear distribution is associated with the spraying gun speed compensation amount, and the scattered point distribution is associated with the viscosity adjustment coefficient.

[0026] Preferably, it further comprises the step S5: detection result credibility verification, a preset number of abnormal areas and normal areas are randomly selected in the heat map, and the selected range of each area covers at least three adjacent grid units;Independent visual evaluation is carried out by at least two detection personnel, and the color difference level determination results are recorded;The consistency index of the model prediction result and the artificial evaluation result is calculated, and the index is quantified by chi-square test method;When the consistency index is lower than the preset threshold, the following processing flow is automatically triggered: freeze the current model parameters, rollback to the last stable version model, and re-execute the online learning process of step S3 until the training error of the new sample recovers to the normal range.

[0027] The beneficial effects of the present application are:

[0028] Through the organic combination of each step, the intelligent color difference analysis detection method for coatings can achieve high precision and intelligent level in coating color difference detection. From multi-spectral image acquisition, texture and color separation, to adaptive modeling and visual analysis, each step helps to solve different challenges in coating color difference analysis. Through real-time data acquisition, texture interference elimination, model adaptive optimization and process defect analysis, the overall not only improves the detection accuracy and efficiency, but also has the characteristics of intelligent adaptation, which can adapt to different production environments and substrate types, thereby optimizing the coating production process and ensuring the consistency of coating quality. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the present application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only a part of the present application, and other drawings can also be obtained by those skilled in the art without any creative effort.

[0030] Fig. 1 Step flow chart of the method of the present application;

[0031] Fig. 2 Step flow chart of the weight update strategy of the online learning algorithm in step S3 of the method of the present application;

[0032] Fig. 3 Step flow chart of the heat map generation method in step S4 of the method of the present application. DETAILED DESCRIPTION

[0033] The present application will be described in detail below in combination with the drawings and specific embodiments. It should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and other alternative ways can also be used by those skilled in the art to implement them; and the drawings are only used to describe the embodiments in more detail, and are not intended to specifically limit the present application.

[0034] Embodiment 1

[0035] See Figs. 1-3 The embodiment of the present application provides an intelligent color difference analysis detection method for coatings. The main purpose of step 1 is to realize the irradiation of the coating surface by the adjustable light source of the ring array, so as to obtain accurate color difference data by multi-band light. By dynamically adjusting the light source intensity, the light source intensity of different regions is adjusted according to the reflectivity reference value obtained by pre-scanning, so as to ensure the uniformity of illumination when image acquisition. The reflected light is separated into different wavebands by a light splitting imaging device and the images are collected synchronously, so as to eliminate the registration error caused by mechanical switching.

[0036] Through the dynamic compensation mechanism, the influence of external lighting conditions is overcome, ensuring the stability and consistency of image data. This process improves the accuracy of color difference analysis, avoiding errors caused by uneven lighting, especially on complex coated surfaces, allowing better capture of real color difference data.

[0037] In step 2, based on image acquisition, texture-related features of the substrate are extracted from the texture-enhanced band images. Using the pre-stored substrate type library, appropriate filter parameters are matched, and a modulation model between texture and color channels is established. This model eliminates the interference of texture on color channels through an iterative optimization algorithm, ultimately generating a pure color layer after decoupling.

[0038] This step can effectively eliminate the interference of substrate texture, ensuring that color difference analysis only focuses on color information, not the influence of surface texture. Through precise texture and color separation, color difference analysis is more accurate, helping to better capture color changes on the coated surface.

[0039] Step 3 matches the pre-trained substrate classification model according to the frequency energy distribution of the texture features, and loads the corresponding initial parameters. During real-time detection, manually reviewed samples are added to the training set, and an online learning algorithm driven by difference is used to dynamically update model weights. Finally, color difference levels and spatial distribution data are output to help detection personnel quickly identify color difference problems.

[0040] Through adaptive modeling, this method can automatically adjust the model according to the characteristics of different coatings, improving the detection ability of coating color difference under various substrates and process conditions. In addition, the introduction of online learning algorithms enables continuous optimization and adaptation to new samples, improving intelligent and adaptive capabilities.

[0041] In step 4, the real-time generated color difference data is mapped to a scalable heat map, and the grid density is dynamically adjusted according to the detection accuracy. This heat map can clearly show the color difference distribution of the coating, helping detection personnel quickly find abnormal areas. By analyzing the geometric feature parameters of abnormal areas and matching them with data in the process defect pattern library, optimized suggestions can be generated and superimposed on the heat map as visual identifiers.

[0042] Through the form of a heat map, color difference data is visually presented, allowing detection personnel to quickly identify and locate color difference problems. In addition, combined with the process defect pattern library, optimized suggestions can be automatically generated, helping production lines adjust process parameters, reducing defects in production, and improving product quality.

[0043] In a possible implementation, in the pre-scanning stage, the detection area is irradiated by a light source at a preset angle with the surface normal of the coating, and the maximum light intensity value in the specular reflection direction and the average light intensity value in the diffuse reflection area are measured using a spectrophotometer. The ratio of the two measured values is calculated as the reflectance reference value. Next, in the real-time detection stage, the detection area is divided into multiple sub-areas, each sub-area is scanned, and the deviation percentage of the reflected light intensity of the sub-area from the reflectance reference value is calculated. When the deviation percentage exceeds a preset threshold, the light source controller generates adjustment instructions according to the deviation direction and amplitude. Specifically, for the area with a reflectance higher than the reference value, the light source intensity is proportionally reduced; and for the area with a reflectance lower than the reference value, the light source intensity is proportionally compensated, so that the deviation of the real-time reflected light intensity from the reference value is controlled within the allowable range.

[0044] Through the specific implementation of this step, the intelligent color difference analysis and detection method for coatings can ensure the uniformity and stability of the light source during image acquisition under various environmental changes and coating characteristics. Dynamic adjustment of the light source intensity not only compensates for errors caused by surface unevenness, but also ensures that the color difference analysis is not affected by inconsistent lighting, thereby providing more accurate detection results and ultimately optimizing the entire coating quality control process.

[0045] In a possible implementation, the reflectance reference value is dynamically corrected according to different types of substrates. Specifically:

[0046] For metal substrates, the correction coefficient is calculated based on the correlation between the surface roughness and the specular reflection light intensity decay rate. The surface roughness is quantified by the gray level gradient variance of the near-infrared image, reflecting the microscopic unevenness of the metal surface. The specular reflection light intensity decay rate represents the change of the reflection ability of the metal surface with the illumination angle.

[0047] For non-metal substrates, the correction coefficient is determined based on the relationship between the texture anisotropy index and the uniformity of the diffuse reflection light intensity distribution. The texture anisotropy index can be calculated by the standard deviation of the texture direction distribution histogram, reflecting the directionality and uniformity of the surface texture of the non-metal substrate.

[0048] The corrected reflectance reference value will be used in the real-time detection stage to adjust the light source intensity, ensuring consistency and comparability of different types of substrates in color difference detection.

[0049] By using the reflectance benchmark value weighting correction method based on the substrate type, the intelligent color difference analysis detection method for coatings can be personalized adjusted according to the lighting and reflection characteristics of different substrates, thereby eliminating the influence of substrate differences on the detection results. This dynamic correction mechanism not only improves the accuracy of color difference analysis, but also enhances the adaptability to different materials, ensuring high quality and high precision of color difference detection in diversified application scenarios. This has significant advantages for coating detection in complex working conditions, especially in industrial production environments that require high precision.

[0050] Embodiment 2

[0051] In one possible implementation, a substrate feature library is first pre-stored, containing texture feature data of typical substrates such as metal, plastic and wood. The data is obtained by taking multi-angle near-infrared imaging of standard samples of each substrate and extracting its texture spatial frequency spectrum and direction distribution histogram. The spatial frequency spectrum reflects the distribution of surface texture in the frequency domain, while the direction distribution histogram represents the distribution of texture in different directions, which are important basis for distinguishing different substrate types.

[0052] The matching condition for metal substrate is that the energy proportion of its spatial frequency spectrum in the preset high frequency interval exceeds the first threshold value. This means that the metal surface has strong high frequency characteristics, usually reflecting a smooth surface and small texture details.

[0053] The matching condition for plastic substrate is that the number of main peaks of its direction distribution histogram is less than the second threshold value, indicating that the plastic surface texture is relatively less and not very complex, mainly showing a relatively uniform or single texture direction.

[0054] The matching condition for wood substrate is that its direction distribution standard deviation is greater than the third threshold value, indicating that the wood surface has a relatively complex and diverse texture direction distribution, with large texture changes and strong directionality.

[0055] During real-time detection, the appropriate filter parameter combination is automatically selected according to the frequency domain features of the current near-infrared image. Specifically, the substrate type of the current image is automatically judged according to the detected spatial frequency spectrum and direction distribution histogram, and the corresponding filter parameters are matched, thereby improving the accuracy of color difference analysis.

[0056] The filter parameter matching method pre-stores substrate feature data and selects appropriate filters according to the frequency domain features of different substrates, so that the intelligent color difference analysis detection method for coatings can accurately adapt to the texture characteristics of various substrates. This not only improves the accuracy of color difference detection, but also enhances the adaptability and automation level in real-time production environment, ultimately realizing an efficient, accurate and intelligent coating detection process.

[0057] In one possible implementation, first, the texture coupling coefficient matrix is initialized to describe the relationship between the texture features and the color channel images. The role of this matrix is to provide the basis for subsequent convolution operations and lay the data foundation for generating the initial texture modulation model.

[0058] The initialized texture feature map is convolved with the color channel images. Through this convolution process, the texture features are coupled with the color information to generate an initial texture modulation model. At this time, the generated model has not been optimized and may contain some residual errors.

[0059] To reduce the errors in the generated texture modulation model, the least squares method is used to iteratively optimize the base layers. The least squares method aims to minimize the mean square error of the texture modulation residuals, making the generated model more accurate and better reflecting the actual color difference.

[0060] In each iteration, the chroma histogram cosine similarity between the current base layer and the previous iteration result is calculated. This step assesses the similarity between the current optimization result and the previous result by analyzing the changes in chroma information, thereby judging the progress of the optimization process.

[0061] When the change in the chroma histogram cosine similarity in the last three iterations is less than the set convergence threshold, the optimization process is considered to have converged, and the iteration is terminated. This judgment depends on the speed of change in the optimization result to determine when to stop iteration, avoiding unnecessary calculations.

[0062] The convergence threshold is dynamically adjusted according to the coating color difference tolerance level. The stricter the coating color difference tolerance level (i.e., the smaller the color difference required), the higher the threshold setting, thereby making the optimization process more refined and ensuring that the final color difference analysis result meets the high-precision requirements.

[0063] The iterative optimization algorithm effectively reduces the errors in the texture modulation model and improves the accuracy of color difference analysis through steps such as initializing the texture coupling coefficient matrix, generating an initial model through convolution, optimizing the base layers using the least squares method, and calculating the chroma histogram cosine similarity. At the same time, the mechanism of dynamically adjusting the convergence threshold enhances flexibility and adaptability, making the method widely applicable to different color difference control precision scenarios, ultimately ensuring the efficiency and accuracy of paint color difference detection.

[0064] In one possible implementation, first, the collected near-infrared image is subjected to a fast Fourier transform (FFT) to convert the image from the time domain to the frequency domain. The frequency domain can more clearly reflect the texture features of the image, especially on the surface features of different materials. The output of the Fourier transform is a frequency energy distribution map, which shows the energy distribution of different frequency components in the image.

[0065] Next, by calculating the energy proportion of the frequency domain energy distribution diagram in the characteristic frequency band of the metal substrate and the non-metal substrate, the frequency domain characteristics of different materials can be obtained. The characteristics of metal and non-metal substrates in the frequency domain are significantly different, and the calculation of energy proportion helps to distinguish between the two different types of substrates.

[0066] The calculated energy proportion data is input into the pre-trained substrate classification model. The classification model learns the boundary conditions of the frequency domain characteristics of different substrates through historical sample data. The substrate classification model judges the substrate type of the current image according to these boundary conditions and performs matching.

[0067] The classification model outputs the substrate type with the highest matching score based on the input frequency domain characteristic data. Based on the matching result, the corresponding pre-trained color difference model parameters are loaded. These model parameters are related to the color difference detection of a specific substrate and can perform accurate color difference analysis according to the substrate type.

[0068] When the matching score is lower than the preset confidence threshold, an artificial review process is triggered. The artificial review process aims to handle situations that cannot be accurately judged to ensure that the final detection result has high reliability. At the same time, the current substrate characteristic data is recorded as new sample data for subsequent continuous optimization and training.

[0069] The pre-trained substrate classification model matching through the steps of Fourier transform to extract frequency domain characteristics, calculation of energy proportion, learning of the classification model, and artificial review process of low matching degree effectively improves the accuracy of substrate identification and the precision of color difference analysis. The combination of automatic processing and artificial review enhances the reliability and flexibility, making the paint color difference analysis and detection process more efficient, accurate, and able to adapt to the needs of various different substrate types.

[0070] In one possible implementation, first, the model is learned online through new sample data. To measure the accuracy of the model's prediction results, the difference between the color difference level of the new sample and the model's prediction result needs to be quantified. This difference is achieved by calculating the Euclidean distance between the two in the three dimensions of hue, saturation, and lightness. As a common measurement method, Euclidean distance can reflect the actual difference between two samples in color space.

[0071] After quantifying the difference between the color difference level and the prediction result by Euclidean distance, the difference obtained is used as the loss function. This loss function represents the error between the model's current prediction and the actual value. Next, the partial derivative of the model's fully connected layer (i.e., the weight layer in the neural network) is calculated based on the loss function, obtaining the gradient direction. The gradient direction indicates how to adjust the weights to minimize the loss function, thereby improving the accuracy of the model.

[0072] To accelerate the learning process and avoid oscillation during gradient descent, the momentum acceleration method is adopted. In this process, in addition to using the current gradient, the gradient update direction of the previous step is also considered for weighted averaging to determine the new weight update step. Momentum acceleration helps to make the convergence process more stable and efficient in the complex optimization of the loss function.

[0073] An automatic adjustment strategy for the learning rate is also introduced. When the loss function of the newly added samples for several consecutive times exceeds the historical average level, the learning rate is automatically expanded. This strategy can accelerate the convergence process of the model, especially when the model has already well fitted the previous samples. By increasing the learning rate, the parameters can be adjusted more quickly, thereby improving the overall training efficiency.

[0074] By introducing online learning algorithms, Euclidean distance loss functions, momentum acceleration methods, and adaptive learning rate expansion strategies, the intelligent color difference analysis and detection method for coatings can continuously optimize in dynamic environments, quickly adapt to new data, and improve the accuracy and efficiency of color difference analysis. These strategies not only enhance real-time performance and accuracy, but also significantly improve the training efficiency and application flexibility of the model, especially when facing a large amount of real-time updated data.

[0075] In one possible implementation, during the heat map generation process, the actual physical size corresponding to a single pixel is first calculated using the spatial resolution of the high-speed image sensor. Spatial resolution generally represents the number of pixels that can be captured per unit area of the image sensor. Through this calculation, the size of each pixel in the actual physical space can be determined. This step provides a basis for subsequent grid cell division density calculation, so that the length of each grid cell matches the actual physical size.

[0076] After calculating the physical size corresponding to each pixel, the grid cell division density needs to be determined according to the preset detection accuracy. Specifically, the length of the grid cell should be equal to an integer multiple of the preset detection accuracy. By reasonably selecting the division density of the grid cell, it is ensured that the detection accuracy meets the expectations, while avoiding excessive computational complexity. This step ensures that each grid cell can reasonably reflect the color difference analysis results without affecting subsequent statistical analysis due to low or high resolution.

[0077] Color difference statistical analysis is performed on all pixel points in each grid cell to calculate the average value and standard deviation of ΔE. ΔE is a standard measure of the difference between two colors, representing the distance between two colors in the color space. In this step, the calculation of ΔE is based on the color information of each pixel, and the average value represents the color difference level of the entire grid cell, while the standard deviation reflects the dispersion degree of color change within the cell.

[0078] After the statistical analysis of color difference is completed, the obtained grid data is usually discrete. In order to make the color difference information more smooth and easy to visualize, a bicubic interpolation algorithm is used to convert the discrete grid data into a continuous color difference distribution heat map. The bicubic interpolation algorithm generates a higher resolution image with smooth transitions by calculating the weighted average of adjacent data points. This step makes the visual effect of the heat map more delicate and the transition of color difference more natural through smoothing processing.

[0079] Finally, the calculated ΔE values are mapped into a heat map through a color gradient mapping, with different ΔE value intervals corresponding to different colors. This gradient mapping helps to visually display the distribution of color difference and clearly identify areas with larger color difference on the heat map. For example, larger ΔE value areas may be displayed in red, while smaller ΔE values are displayed in green or blue. Color gradient mapping allows users to quickly identify parts of the paint surface with larger color difference, facilitating further processing or optimization.

[0080] By combining color difference statistical analysis, interpolation algorithm and color gradient mapping, the heat map generation method can effectively convert the color difference of the paint surface into an intuitive image. This method not only improves the accuracy and efficiency of color difference analysis, but also helps users quickly identify problem areas by optimizing visual effects, providing strong support for quality control and production optimization of paint.

[0081] In one possible implementation, first, color difference distribution samples from historical production need to be collected through devices and sensors. These samples mainly come from common problems that occur during the spraying process, such as abnormal spraying pressure, deviation of spraying gun movement speed and improper paint viscosity. These abnormal factors may cause color difference in the production process, thereby affecting the quality of the final paint. By collecting these samples, color difference data under different process conditions can be obtained to provide data support for subsequent analysis and model construction.

[0082] Each collected color difference distribution sample is analyzed and the following geometric features are extracted:

[0083] Area proportion of abnormal area: represents the proportion of the area with larger color difference in the entire image, reflecting the concentration of color difference distribution.

[0084] Centroid position offset: calculates the offset of the centroid of the area with larger color difference from the image centroid, which can reflect the influence of some uneven factors in the spraying process.

[0085] Edge sharpness index: measures the sharpness of the color difference distribution edge. A too fuzzy edge may indicate unstable movement speed of the spraying gun or viscosity problems in the spraying process.

[0086] The extracted geometric features are analyzed and classified using clustering algorithms, dividing different color difference distribution samples into three typical distribution patterns:

[0087] Ring distribution: usually related to too low spraying pressure or uneven pressure during spraying. This distribution often presents large color difference areas in a ring or near-circular shape in the image.

[0088] Linear distribution: commonly caused by spraying gun movement speed deviation. This distribution presents a linear shape, usually due to too fast or too slow spraying gun speed, leading to uneven paint spraying.

[0089] Scattered point distribution: mainly caused by paint viscosity problems. Too high or too low paint viscosity can cause uneven small spots during spraying.

[0090] For each distribution pattern, a corresponding process parameter adjustment rule library is established:

[0091] For ring distribution, reduce the area of color difference regions by lowering the spraying pressure to improve paint distribution uniformity.

[0092] For linear distribution, adjust the compensation amount of the spraying gun movement speed to reduce the color difference of linear distribution, thereby optimizing the uniformity of spraying.

[0093] For scattered point distribution, adjust the viscosity of the paint to improve color difference distribution, making the paint flow more stable and avoiding scattered point phenomenon.

[0094] By collecting historical production data, extracting geometric features, applying clustering algorithms and establishing process parameter adjustment rule library, accurate identification and optimization path are provided for color difference problems in paint production process. Through this data-driven method, the quality control capability of paint production can be significantly improved, process parameters can be optimized, waste rate can be reduced, and production efficiency can be improved.

[0095] In one possible implementation, in the heat map, first randomly select a predetermined number of regions, including abnormal regions and normal regions. The selection range of these regions should cover at least three adjacent grid cells. The selected regions include abnormal regions with large color difference and normal regions with small or no obvious color difference, ensuring comprehensive evaluation of the model prediction results.

[0096] For the selected abnormal and normal regions, at least two inspectors will independently conduct visual evaluation for each region. The inspectors will grade the color difference of each region according to the visual perception of color difference. This process mainly relies on human experience to compare with the model prediction results and evaluate the accuracy of the model.

[0097] Each inspector independently records their color difference level determination results for each selected area. These determination results will be used for subsequent consistency analysis to quantify the difference between model prediction results and manual evaluation.

[0098] To quantify the consistency of model prediction and manual evaluation results, a consistency index needs to be calculated. This index is quantified by the chi-square test method. The chi-square test method can be used to test the deviation between the observed data and the expected data, and then obtain the consistency index, which represents the consistency degree of the model prediction results and the manual evaluation results.

[0099] If the consistency index is lower than the preset threshold, it means that there is a large difference between the model prediction results and the manual evaluation results, and at this time the automatic processing process needs to be triggered:

[0100] Freeze the current model parameters: stop further optimization of the current model to avoid continuing to use the model in an unstable state.

[0101] Roll back to the last stable version of the model: restore to the previous verified and stable model version to ensure that the model used in the production process can achieve the expected performance.

[0102] Re-execute the online learning process of step S3: go back to step S3 and continue online learning to train a new model until the training error of the new sample returns to the normal range. This process helps to optimize the model so that it can more accurately predict color difference problems.

[0103] By verifying the consistency between model prediction and manual evaluation, the accuracy and reliability of the paint color difference detection results are ensured. By quantifying the consistency index through chi-square test and automatically rolling back to the stable version when necessary, continuous optimization and iteration can be carried out to improve the intelligent degree of the detection process, reduce manual intervention, and at the same time ensure the control of production quality.

[0104] Example 3

[0105] The application of the present application in the intelligent color difference detection of automobile body spraying line specifically includes:

[0106] Application scenario: a certain automobile manufacturing plant coating workshop, color difference detection after spraying epoxy primer on cold rolled steel plate (Ra=0.8-1.2μm), requiring ΔE≤0.8.

[0107] Step S1: multispectral dynamic compensation image acquisition

[0108] 1. Light source parameters:

[0109] Ring array 12 groups LED light source (center wavelength 450 nm, 550 nm, 650 nm), with the normal of the steel plate at 45° angle, initial intensity 1200 lx.

[0110] 2. Reflectivity reference value:

[0111] The maximum light intensity of the specular reflection in the pre-scanning left front door area is 1850 lx, and the average light intensity of the diffuse reflection is 320 lx, and the reference value R = 1850 / 320 = 5.78.

[0112] 3. Real-time adjustment result:

[0113] The reflectivity deviation of the left front door area is +7.2%, and the light source intensity is lowered to 1100 lx; the reflectivity deviation of the right rear wheel arch area is -4.5%, and the light source intensity is increased to 1250 lx.

[0114] The final reflectivity deviation of each area is controlled within ±2%.

[0115] Step S2: Texture-color coupling separation

[0116] 1. Filter matching:

[0117] The near-infrared image (940 nm) frequency domain analysis shows that the high-frequency energy of the metal substrate accounts for 82% (threshold 80%), and a high-pass filter (cutoff frequency 1.2 mm⁻¹) is loaded.

[0118] 2. Iterative optimization result:

[0119] The initial chroma histogram cosine similarity is 0.72, which is improved to 0.98 after 5 iterations (convergence threshold ΔHSL <0.01).

[0120] The output decoupling color layer has a hue deviation of ≤0.3° and a brightness deviation of ≤0.5%.

[0121] Step S3: Adaptive color difference modeling

[0122] 1. Substrate classification:

[0123] The FFT frequency energy distribution shows that the metal characteristic frequency band (0.5-2 mm⁻¹) accounts for 91%, and the pre-trained metal substrate classification model is matched (confidence 98.7%).

[0124] 2. Online learning result:

[0125] Manual review of 10 samples, loss function (ΔE Euclidean distance weight 0.6:0.3:0.1) calculates the average gradient descent step size 0.02, momentum coefficient 0.9.

[0126] After training the new samples, the model prediction error rate decreased from 2.1% to 0.9%.

[0127] Step S4: Process correlation visualization

[0128] 1. Heat map generation:

[0129] Image resolution 0.1 mm / pixel, grid density 0.5 mm, ΔE average 0.65 (standard deviation 0.12), using HSV gradient (ΔE = 0 → red, ΔE = 1.0 → blue).

[0130] 2. Defect analysis results:

[0131] The left front door local area (area ratio 6.2%) presents a ring-like distribution, triggering the spray pressure compensation instruction (pressure from 0.45 MPa to 0.42 MPa).

[0132] The area with edge sharpness index > 25° (ratio 3.8%) is associated with spray gun speed compensation (+15 cm / s).

[0133] Step S5: Detection result credibility verification

[0134] 1. Artificial review results:

[0135] Randomly select 20 abnormal points (covering 6 grid units), 5 inspectors visually determine 12 samples with ΔE > 0.8, consistent rate with model results 91.7%.

[0136] 2. Model update results:

[0137] Consistency index χ² = 0.032 (α = 0.05), no rollback mechanism triggered, current model version retained.

[0138] Final output:

[0139] Color difference level distribution chart: qualified rate 98.3% (ΔE ≤ 0.8), ΔE > 0.8 area marked in red (total 3, area ratio 0.7%).

[0140] Process optimization suggestions:

[0141] Left front door spray pressure: 0.42 MPa (original 0.45 MPa);

[0142] Spray gun speed compensation: +15 cm / s (left front door), +8 cm / s (right rear wheel arch).

[0143] The present application encompasses any alternatives, modifications, equivalent methods and solutions made to the essence and scope of the present application. In order to make the public have a thorough understanding of the present application, specific details are described in the following preferred embodiments of the present application, and the present application can also be fully understood without the description of these details to those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, well-known methods, processes, procedures, elements and circuits, etc. are not described in detail.

[0144] The above is only the preferred embodiment of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can also be made, which should be considered as the protection scope of the present application.

Claims

1. A method for intelligent color difference analysis for coatings, characterized in that, The method comprises the following steps: Step S1: multi-spectral dynamic compensation image acquisition, in a closed detection cabin, a multi-band light is projected to a coating surface by an adjustable light source of a ring array, the intensity of each region light source is dynamically adjusted according to the reflectivity reference value obtained by pre-scanning; the reflected light is separated into a human eye sensitive band and a texture enhancement band by using a spectral imaging device, multi-channel image data is synchronously captured, and registration errors caused by mechanical switching are eliminated; the reflectivity reference value is dynamically calculated by the intensity ratio of diffuse reflection and specular reflection; Step S2: texture-color coupling separation, substrate-related texture features are extracted from the texture enhancement band image, filter parameter matching filter parameters are matched based on a pre-stored substrate type library, multi-scale convolution operation is performed on the texture enhancement band image by using the filter, and a frequency domain response matrix containing substrate inherent texture features is extracted; the frequency domain response matrix containing substrate inherent texture features is tensor fused with an RGB color channel, a texture modulation model of the color channel is established, texture interference is eliminated by using an iterative optimization algorithm, and a decoupled pure color layer is generated; the iterative optimization takes the similarity of adjacent iteration chroma histograms as a termination condition; Step S3: adaptive chromatic aberration modeling, a pre-trained substrate classification model is matched according to the texture feature frequency energy distribution, and initial parameters are loaded; A manual review sample in real-time detection is added to a training set, a difference degree driven online learning algorithm is used to dynamically update model weights, and chromatic aberration grades and spatial distribution data are output; Step S4: process correlation visualization, chromatic aberration data is mapped into a scalable heat map, the grid density is dynamically adjusted according to detection accuracy, geometric feature parameters of an abnormal area are analyzed, optimization suggestions are generated by matching a process defect mode library, and the optimization suggestions are superimposed on the heat map in a visualized identification.

2. The intelligent color difference analysis and detection method for paint according to claim 1, characterized in that, The specific implementation method for dynamically adjusting the light source intensity in the step S1 comprises the following steps: In the pre-scanning stage, a light source at a preset angle with the normal line of the coating surface irradiates the to-be-detected area, the maximum light intensity value in the specular reflection direction and the average light intensity value in the diffuse reflection area are measured by a spectrophotometer respectively, the ratio of the two is taken as the reflectivity reference value; in real-time detection, the current detection area is scanned in blocks, the deviation percentage of the reflected light intensity of each sub-region from the reference value is calculated, when the deviation percentage exceeds a preset threshold, the light source controller generates an adjustment instruction according to the deviation direction and amplitude: for the area with a reflectivity higher than the reference value, the light source intensity at the corresponding position is reduced in proportion; for the area with a reflectivity lower than the reference value, the light source intensity is compensated in proportion, until the deviation of the real-time reflected light intensity from the reference value is controlled within the allowable range.

3. The intelligent color difference analysis and detection method for paint according to claim 2, characterized in that, The weighted correction method of the reflectivity reference value comprises the following steps: The reference value is dynamically corrected according to the coating substrate type, the correction coefficient of the metal substrate is calculated through the correlation between the surface roughness and the specular reflection light intensity decay rate, and the correction coefficient of the non-metal substrate is determined through the relationship between the texture anisotropy index and the diffuse reflection light intensity distribution uniformity; the roughness is quantified by the gray gradient variance of the near-infrared image, and the anisotropy index is calculated by the standard deviation of the texture direction distribution histogram.

4. The intelligent color difference analysis and detection method for paint according to claim 1, characterized in that, The parameter matching method of the filter in the step S2 comprises the following steps: The pre-stored substrate feature library stores typical texture feature data of metal, plastic and wood, which is obtained by multi-angle near-infrared imaging of standard samples of each substrate, and extracting the texture spatial frequency spectrum and direction distribution histogram; the determination condition of the metal substrate matching the high-frequency filter is that the energy ratio of the spatial frequency spectrum in the preset high-frequency interval exceeds the first threshold value, the determination condition of the plastic substrate matching the medium-frequency filter is that the number of main peaks of the direction distribution histogram is less than the second threshold value, and the determination condition of the wood substrate matching the multi-direction filter is that the standard deviation of the direction distribution is greater than the third threshold value; during real-time detection, the filter parameter combination is automatically selected according to the frequency domain features of the current near-infrared image.

5. The intelligent color difference analysis and detection method for paint according to claim 1, characterized in that, The implementation process of the iterative optimization algorithm in step S2 includes: Initializing the texture coupling coefficient matrix, performing convolution operation on the texture feature map and the color channel image to generate an initial texture modulation model; the base layer is iteratively optimized by the least square method to minimize the mean square error of the texture modulation residual; in each iteration, the cosine similarity of the chroma histogram of the current base layer and the previous iteration result is calculated, and the optimization is terminated when the similarity changes by less than a convergence threshold value for three consecutive iterations; the convergence threshold value is dynamically adjusted according to the color difference tolerance level of the coating, and the higher the color difference requirement, the higher the threshold value.

6. The intelligent color difference analysis and detection method for paint according to claim 1, characterized in that, The detailed process of matching the pre-trained substrate classification model in step S3 includes: Performing fast Fourier transform on the near-infrared image to obtain a frequency energy distribution graph, and calculating the energy ratio in the metal substrate feature frequency band and the non-metal substrate feature frequency band; input the energy ratio into the pre-trained substrate classification model, and the classification model learns the frequency domain feature boundary conditions of different substrates through historical sample data; output the substrate type with the highest matching degree score, and load the corresponding pre-trained color difference model parameters; when the matching degree score is lower than the confidence threshold value, trigger the manual review process and record the new substrate feature data.

7. The intelligent color difference analysis and detection method for paint according to claim 1, characterized in that, The weight update strategy of the online learning algorithm in step S3 includes: Quantifying the difference between the color difference level of the new sample and the model prediction result into a loss function, the difference is determined by calculating the Euclidean distance in hue, saturation and lightness dimensions; calculate the gradient direction according to the partial derivative of the loss function to the weight of the model full connection layer, and adjust the weight update step using momentum acceleration method; when the loss function of a plurality of new samples in succession decreases at a rate exceeding the historical average level, automatically expand the learning rate to accelerate model convergence.

8. The intelligent color difference analysis and detection method for paint according to claim 1, characterized in that, The heat map generation method in step S4 includes: According to the spatial resolution of the high-speed image sensor, the actual physical size corresponding to a single pixel is calculated, the division density of the grid unit is determined, so that the length of each unit is equal to an integer multiple of the preset detection accuracy; perform color difference statistical analysis on all pixel points in each grid unit, and calculate the average value and standard deviation of ΔE value; use the bicubic interpolation algorithm to convert the discrete grid data into a continuous color difference distribution heat map, and use color gradient mapping to reflect different ΔE value intervals. ΔE is a standard measure of color difference, representing the distance between two colors in the color space.

9. The intelligent color difference analysis and detection method for paint according to claim 1, characterized in that, The method for constructing the process defect mode library in the step S4 comprises the following steps: Collecting color difference distribution samples caused by abnormal spraying pressure, spraying gun moving speed deviation, and improper paint viscosity in historical production, and extracting the following geometric features for each sample: area proportion of abnormal area, center of gravity position offset, and edge sharpness index; the feature data is divided into three modes of ring distribution, linear distribution, and scattered point distribution through a clustering algorithm; a process parameter adjustment rule library is established for each mode, the ring distribution is associated with a spraying pressure reduction ratio, the linear distribution is associated with a spraying gun speed compensation amount, and the scattered point distribution is associated with a viscosity adjustment coefficient.

10. The intelligent color difference analysis and detection method for paint according to claim 1, characterized in that, The step S5 further comprises a detection result credibility verification, a preset number of abnormal areas and normal areas are randomly selected in the heat map, and the selection range of each area covers at least three adjacent grid units; Independent visual evaluation is performed by at least two detection personnel, and color difference grade determination results are recorded; a consistency index of the model prediction result and the artificial evaluation result is calculated, the index is quantified through a chi-square test method; when the consistency index is lower than a preset threshold value, the following processing procedure is automatically triggered: freezing the current model parameter, returning to the last stable version model, and re-executing the online learning process of the step S3 until the training error of the new sample is restored to a normal range.

Images