CN-122023346-A - Paint surface roughness nondestructive detection method and system for industrial machine vision

CN122023346ACN 122023346 ACN122023346 ACN 122023346ACN-122023346-A

Abstract

The invention discloses a paint surface roughness nondestructive testing method and system for industrial machine vision, which belong to the technical field of visual detection and comprise the following steps of selecting a group of standard paint surface test blocks with known real roughness values, carrying out image acquisition under different illumination parameter combinations, defining illumination parameters as file names of images to form an image training set, preprocessing the image training set, extracting global gray average values in the preprocessed images, converting the illumination parameters of the preprocessed images into illumination parameter vectors, inputting the illumination parameter vectors, the real roughness values and the global gray average values into a roughness-characteristic relation model to train the roughness-characteristic relation model, and utilizing the trained roughness-characteristic relation model. The invention adopts non-contact optical imaging to ensure that no physical damage is caused to the paint surface.

Inventors

ZHOU ROUGANG
XU QINGYANG
YU JIAHUI
ZHAO SHUWEN
ZENG DONGSHENG
CHEN AN
ZHOU CAIJIAN
CHEN ENZAN

Assignees

杭州汇萃智能科技有限公司
苏州汇萃智能科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260131

Claims (10)

1. The paint surface roughness nondestructive testing method for industrial machine vision is characterized by comprising the following steps of: Step one, selecting a group of standard paint surface test blocks with known real roughness values; Step two, image acquisition is carried out under different illumination parameter combinations, illumination parameters are defined as file names of images, and an image training set is formed; preprocessing an image training set, extracting a global gray average value in a preprocessed image, converting illumination parameters of the preprocessed image into illumination parameter vectors, and inputting the illumination parameter vectors, a real roughness value and the global gray average value into a roughness-characteristic relation model to train the roughness-characteristic relation model; And fourthly, predicting the roughness of the paint surface with unknown roughness by utilizing the trained roughness-characteristic relation model, judging whether the predicted result is smaller than a set threshold value, if so, judging that the paint surface is qualified, and if not, judging that the paint surface is unqualified.
2. The method of claim 1, wherein the illumination parameter comprises an incident angle of a light source Distance of light source from center And light source height 。
3. The detection method according to claim 2, wherein the preprocessing of the training set of images is performed as follows: Step 1, correcting an original image; step 2, carrying out enhancement processing on the image after flat field correction by adopting limited contrast self-adaptive histogram equalization; step 3, carrying out relative reflectivity normalization processing based on a reference white board on the enhanced image; And 4, filtering the normalized image by using a rotation invariance Gabor filter bank or anisotropic diffusion filtering to obtain a preprocessed image.
4. A detection method according to claim 3, wherein the extraction of the global gray average value from the preprocessed image is performed as follows: step 1, decoding the preprocessed image into a three-dimensional array, wherein the three-dimensional array comprises a height Width of the container Sum channel number Number of channels 3, Respectively red Green and environment-friendly And blue color ; Step 2, obtaining a two-dimensional matrix of gray values through gray values of each pixel in the three-dimensional array of the gray calculation formula; and 3, calculating a global gray average value through gray values of the two-dimensional matrix.
5. The method according to claim 4, wherein the first step Gray value of each pixel The calculation is as follows: ; In the color of red, the color of the light is red, Is in the form of green color, and the color is green, Is blue.
6. The method of claim 5, wherein the global gray level average value The calculation is as follows: ; ; the total number of pixels in the preprocessed image.
7. The method of claim 6, wherein the roughness-feature relation model is calculated as follows: ; In order to predict the roughness value(s), 、、 And In order to fit the coefficients of the coefficients, Is an intercept term.
8. The method of claim 2, wherein the roughness-feature relation model training operation is as follows: step 1, lighting parameter vector and true roughness value The global gray average value is input into a roughness-characteristic relation model, and a predicted roughness value is calculated; Step 2, calculating a predicted error value of each group of real roughness values and predicted roughness values; Step 3, square adding the prediction error values of all images to obtain total error Step 4, finding out the minimum total error by a matrix operation method 、、、 And b value until the mean square error of all images is less than the set value.
9. The method according to claim 2, wherein the specific steps of step 4 are as follows: Step 41 constructing a matrix Vector of parameters And a true value vector ; Step 42 solving the normal equation to find the optimal parameter vector Optimal parameter vector In the inner part 、、、 And b minimizes the total error.
10. Paint surface roughness nondestructive test system towards industrial machine vision, its characterized in that includes: The illumination light supplementing module is used for supplementing light during image acquisition and adjusting the incident angle of the light source according to the requirement; the image acquisition module is used for acquiring images; the dark field and positioning module is used for shading by the image acquisition module and the illumination light supplementing module and determining the distance from the light source to the center and the height of the light source; the processing and control module is used for controlling the illumination light supplementing module and the image acquisition module, extracting image characteristics, carrying out image roughness through the extracted characteristics, and calculating the roughness of the paint surface according to the roughness-characteristic relation model.

Description

Paint surface roughness nondestructive detection method and system for industrial machine vision Technical Field The invention relates to the technical field of visual detection, in particular to a paint roughness nondestructive detection method and system for industrial machine vision. Background In high-end manufacturing industries, such as automotive coating, consumer electronics housing treatment, furniture surface treatment, etc., the surface quality of a paint coating is a key indicator that determines the appearance grade and durability of a product. Wherein, the roughness (or texture degree, orange peel effect) of the paint surface is a core measurement parameter; At present, in order to detect the roughness of the paint surface in the industry more rapidly, a qualitative or semi-quantitative evaluation scheme based on machine vision is designed: qualitative or semi-quantitative evaluation based on machine vision, generally using an industrial camera in combination with a uniform light source, photographing the paint surface, and then extracting features through an image processing algorithm. Common methods include: And (5) gray statistics, namely calculating the overall gray mean value, variance, entropy and the like of the image. The method is simple and quick, is extremely sensitive to illumination change, cannot effectively distinguish texture caused by roughness from change caused by irrelevant factors such as color, stain and the like, and has poor robustness. Texture analysis algorithms, such as gray level co-occurrence matrices, characterize texture by computing the spatial relationship of pixel pairs in an image. The method is more advanced than simple gray level statistics, but has high feature dimension, relatively complex calculation and the effect thereof is seriously dependent on the uniformity and stability of illumination. In a practical factory environment, it is difficult and costly to ensure absolute uniformity of global illumination. And a deep learning model, which is to train a large number of marked paint images by using a convolutional neural network and learn the mapping from the images to roughness values. This approach has great potential and can learn complex features. But its performance is highly dependent on the amount and quality of the training data. Obtaining a large number of precisely marked paint samples (i.e., each sample requires prior measurement of its true roughness with contact or high precision optics) is costly and long-lived. Furthermore, models have uncertainty in their generalization ability for new colors, new substrates, or new lighting conditions. Based on the method and the system, the invention designs a paint surface roughness nondestructive testing method and a paint surface roughness nondestructive testing system for industrial machine vision so as to solve the problems. Disclosure of Invention Aiming at the defects existing in the prior art, the invention provides a paint surface roughness nondestructive testing method and system for industrial machine vision. In order to achieve the above purpose, the invention is realized by the following technical scheme: The nondestructive testing method for the roughness of the paint surface facing to the vision of the industrial machine comprises the following steps: Step one, selecting a group of standard paint surface test blocks with known real roughness values; Step two, image acquisition is carried out under different illumination parameter combinations, illumination parameters are defined as file names of images, and an image training set is formed; preprocessing an image training set, extracting a global gray average value in a preprocessed image, converting illumination parameters of the preprocessed image into illumination parameter vectors, and inputting the illumination parameter vectors, a real roughness value and the global gray average value into a roughness-characteristic relation model to train the roughness-characteristic relation model; And fourthly, predicting the roughness of the paint surface with unknown roughness by utilizing the trained roughness-characteristic relation model, judging whether the predicted result is smaller than a set threshold value, if so, judging that the paint surface is qualified, and if not, judging that the paint surface is unqualified. Further, the illumination parameter includes an incident angle of the light sourceDistance of light source from centerAnd light source height。 Further, the specific operation of preprocessing the image training set is as follows: Step 1, correcting an original image; step 2, carrying out enhancement processing on the image after flat field correction by adopting limited contrast self-adaptive histogram equalization; step 3, carrying out relative reflectivity normalization processing based on a reference white board on the enhanced image; And 4, filtering the normalized image by using a rotation invariance Gabor filter bank o