Furniture spraying quality detection method

Abstract

The invention discloses a furniture spraying quality detection method, which comprises the following steps of: 1, acquiring a surface coating image of furniture on a building material by using an industrial camera, and preprocessing the image; 2, carrying out feature extraction on the preprocessed collected image, and obtaining a corresponding building material surface feature map according to the extracted features; and 3, combining the results of the coating areas obtained in the step 1 with different labels to serve as the input of a detection model, and evaluating the furniture spraying quality according to the output result of the network model. According to the invention, accurate detection of the furniture spraying quality on the surface of the building material on an assembly line can be accelerated, and the building material can be better produced and processed according to the spraying quality.

Description

Technical Field

[0001] The present invention relates to the application field of computer vision in the field of furniture and provides a method for detecting the spraying quality of furniture. Background Art

[0002] The rough surface of furniture increases the contact surface area with the coating, improving the coating's adhesion to the building material. However, excessive roughness can easily lead to insufficient coating thickness at wave crests, causing early pitting corrosion. It can also trap bubbles in deeper pits, affecting the furniture's coating quality. Furthermore, traditional target detection methods struggle to achieve high-precision results when inspecting the coating quality of furniture on building material surfaces. Summary of the Invention

[0003] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a furniture spray quality detection method, which includes the following steps: Step 1: Use an industrial camera to obtain the surface coating image of the furniture on the building material, and preprocess the image; Step 2: Extract features from the preprocessed captured image, and obtain the corresponding building material surface feature map based on the extracted features; Step 3: Combine the coating area results obtained in the above steps with different labels as the input of the detection model, and evaluate the quality of the furniture spray quality according to the output results of the network model.

[0004] Furthermore, the step one is specifically as follows: a CMOS camera is placed above the processing line of the spray furniture process to collect surface images of building materials, noise reduction processing is performed on the collected surface images of building materials, and a ring light source is installed under the camera to obtain a pre-processed surface coating image.

[0005] Furthermore, the step 2 is specifically as follows: after acquiring the surface coating image formed by spraying the furniture, the local changes in the surface coating image are first obtained, and a certain number of grid forms are used as comparison objects when obtaining local features, thereby increasing image information and eliminating the influence of grid edgeization.

[0006] Furthermore, the step three is specifically as follows: according to the above steps, the area containing inferior coating in the surface coating image is obtained, the minimum bounding rectangle of these areas containing inferior coating is obtained, the position of inferior coating is located in the surface coating image, different labels are assigned to the minimum bounding rectangle, and the final spraying quality of furniture is detected through the classification network. The size of the minimum bounding matrix may be different, and size normalization processing is required before input into the neural network. The label of the neural network is annotated with the help of the Labelme tool. Label 0 corresponds to the sagging phenomenon in the inferior coating, label 1 corresponds to the bottom leakage phenomenon in the inferior coating, and label 2 corresponds to the bottom leakage phenomenon in the inferior coating. Label 2 corresponds to the bubble phenomenon in inferior coating, and label 3 corresponds to the dirty spot phenomenon in inferior coating. The output of the neural network corresponds to different phenomena of inferior coating in the surface coating image. The spraying quality of furniture is evaluated based on whether inferior spraying occurs after the furniture is sprayed and how many types of inferior spraying phenomena occur. The larger the area of ​​inferior coating area on the surface of a building material and the more types of inferior phenomena appear, the inferior quality of the spraying of this building material furniture. Such building materials need to be removed from the assembly line for re-spraying of the furniture. If there is no inferior coating area on the surface of the building material, it means that the quality of the spraying of the building material furniture is high-quality.

[0007] Furthermore, the step 2 is specifically as follows: assuming that the size of the preprocessed surface coating image is M*N, the surface coating image is divided into m small areas of the same size, the empirical value of m is 200, and any small area is divided into a*a small grids, a takes an empirical value of 5, and each small area is divided into 5*5. In order to extract as much local image information as possible without generating information redundancy, a sliding window is set with a window size of 3*3. The window slides in the small area with a step size of 1 small grid without repetition. Each sliding result is regarded as a coating image composition. Traversing the entire small area will obtain multiple coating image compositions, and there is no repetition between the coating image compositions; the coating image composition contains part of the image information, and the similarities and differences between the image information contained in the pixels within the image composition reflect whether the current area contains different spraying effects. Here, a coating difference factor is constructed, which is used to characterize the grayscale difference between the pixels within a coating image composition. By comparing the coating difference factors of the coating image composition at the center position and the surrounding adjacent coating image compositions in each 5*5 small area, the local change characteristics that can characterize each small area are extracted from the comparison results.

[0008] Furthermore, the multidimensional features of each coating factor are calculated, and all small areas on the entire surface coating image are traversed. The coating difference factors are updated according to the above steps, and the change amount of each area is calculated. Here, the coating inferior feature is constructed to characterize the local changes in the surface coating image. The local changes are manifested in the surface coating image as obvious edge changes and unstable gradient changes between the inferior spraying area and the background area; the coating inferior feature I reflects the difference in image information in the horizontal and vertical directions within the small area, and is also the difference between the quality of spraying in the small area. The larger the coating inferior feature is, the more likely this small area is to contain an inferior coating area; traverse the entire surface coating image. A surface coating image is obtained, and the coating inferior feature I corresponding to each small area is obtained. A total of m*n coating inferior features I are obtained for the surface coating image. All small areas are mapped one by one to pixels to obtain the feature map F corresponding to the surface coating image. The coating inferior feature I value is mapped to the grayscale value of the pixel. The range of the mapped value is [0,.255]. The effective features in the feature map F are very sparse. Consider using a more compact feature to represent the changes in the pixel points in the feature map F. Use a 5*5 window size to traverse the entire feature map F without repetition. For each traversal of the window area, obtain the gradient amplitude and gradient direction of the pixel point. Further, calculate the gradient histogram of the pixels in the area, with a division interval of 20°, and obtain the gradient histogram.

[0009] This paper proposes a computer vision-based furniture spray-coating quality inspection method. By analyzing local variations and global information in captured building material surface images, high-dimensional features are captured and then further reduced to produce corresponding surface feature maps. Next, the label-encoded surface feature maps are used as input to a neural network. The neural network's learning capabilities are leveraged to accelerate the accurate inspection of furniture spray-coating quality on building material surfaces on an assembly line. This allows for better control of building material production and processing based on the quality of the spray coating. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A coating difference factor processing flow chart provided in an embodiment of the present application.

[0011] Figure 2 A histogram provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The present invention is further described below with reference to the accompanying drawings and examples.

[0013] Example: Step 1: Use an industrial camera to obtain an image of the surface coating of furniture on building materials and pre-process the image.

[0014] A CMOS camera is placed above the processing line of the spray furniture process to collect surface images of building materials. The CMOS camera has the characteristics of high integration, fast readout, low power consumption, etc., and is suitable for industrial image collection. The output image of the CMOS camera is an RGB image. Due to the fixed noise of the internal components of the CMOS camera and the interference of the surrounding environment, it is necessary to perform noise reduction processing on the collected surface images of building materials to improve the image quality. On the other hand, building materials generally have a certain reflective ability to light sources. In order to reduce the impact of light on the collected images, a ring light source is installed under the camera. In the field of computer vision, commonly used denoising techniques include bilateral filtering denoising, median filtering denoising, Gaussian filtering denoising, etc. In the present invention, the median filtering denoising technology is used to eliminate the impact of noise in the process of collecting images. Median filtering denoising is a well-known technology, and the specific process will not be described in detail.

[0015] At this point, the pre-processed surface coating image is obtained.

[0016] Step 2: Extract features from the pre-processed surface coating image and obtain the corresponding building material surface feature map based on the extracted features.

[0017] After acquiring the surface coating image formed by spraying furniture, the present invention considers evaluating the local changes in the surface coating image and determining the global information features. The first thing to obtain is the local changes in the surface coating image. Because the areas with poor spraying quality are relatively concentrated in a certain part of the building material, the local changes reflect the reasons for the quality of the spraying. The four common poor spraying phenomena in the spraying process of building materials produced on the same assembly line are sagging, leakage, bubbles, and dirty spots. The determination of global information features is the second focus of the present invention on surface coating images. Global information features reflect the quality of spraying on an entire surface coating image. It refers to the total area of ​​poor spraying quality areas and the number of poor quality phenomenon categories detected in a surface coating image.

[0018] Four types of inferior spraying often occur in furniture during the spraying process of building materials: sagging, leakage, bubbles, and dirty spots. Observing the surface coating images of the above four phenomena, it can be seen that the areas where these phenomena occur have obvious gradient changes, and the edge gradient directions of areas with different inferior phenomena are not consistent.

[0019] In the traditional LBP feature extraction process, binary LBP features are obtained by comparing the grayscale values ​​of neighboring pixels and the central pixel. However, due to the presence of independent noise points, the local LBP features obtained by comparing adjacent pixels are easily disturbed, and the description of local changes in the image is not accurate enough. In the present invention, a grid containing a certain number of grids is used as the comparison object when obtaining local features. It is worth noting that when dividing the surface coating image into grids, the phenomenon of grid marginalization may occur. Therefore, it is necessary to consider adding image information to eliminate the influence of grid marginalization when obtaining local features.

[0020] Assume that the size of the preprocessed surface coating image is M*N, and the surface coating image is divided into m small areas of the same size, with an empirical value of m being 200. Any small area is divided into a*a small grid, with an empirical value of a being 5. At this time, each small area is divided into 5*5. In order to extract as much local image information as possible without generating information redundancy, a sliding window is set with a window size of 3*3. The window slides in the small area without repetition with a step size of 1 small grid. Each sliding result is composed of a coating image. Traversing the entire small area will yield The coating image composition contains part of the image information. The differences and similarities between the image information contained in the internal pixels of the image composition reflect whether the current area contains different spraying effects. The coating difference factor is constructed here. The coating difference factor is used to characterize the grayscale difference between the internal pixels of a coating image composition. The coating difference factor of the coating image composition is calculated. :

[0021]

[0022] Where, and are the maximum and minimum grayscale values ​​within the coating component factor, V is the total number of pixels within the coating component factor, is the grayscale value of any pixel within the coating component factor, is the mean gray value of the coating component factors.

[0023] Traversing all coating image components in a small area can obtain 9 coating difference factors, which are recorded as , arrive , The coating difference factor corresponding to the center position. Compare the coating difference factors of the center position coating image composition and the surrounding adjacent coating image composition in each 5*5 small area, and extract the local change features that can characterize each small area from the comparison results. In this scheme, the multidimensional features of each coating factor are calculated in order to obtain more image information. The multidimensional feature extraction process is as follows: Figure 1 As shown,

[0024] First, compare the coating difference factor of the coating image at the center with the coating difference factor of the surrounding area, and update the coating difference factor in each 5*5 small area according to the size relationship. The specific update process is as follows:

[0025]

[0026] Where i is any one of the eight adjacent coating difference factors. If , then Updated to the minimum of the 9 coating difference factors , if satisfied , then The value of is updated to the minimum value among the 9 coating difference factors .

[0027] Get the new coating difference factor and obtain the change in each area before and after the update If the small area corresponds to a high-quality coating area, then the nine coating difference factors are all small values ​​that are very close in size. This is because the high-quality coating area has consistent color and uniform brightness, and the image information carried by each pixel is relatively close. If the regional variation is large, it means that the coating difference factors in this area are quite different. Correspondingly, the image information in each 3*3 coating image composition has obvious differences.

[0028]

[0029] Where p and j are the coating image components with different coating difference factors before and after updating. 、 They are the number of coating difference factors less than the center position and the number of coating difference factors greater than or equal to the center position, respectively. It reflects the information difference between different coating image compositions in a small area. The larger the value of , the more likely this small area is to contain an area of ​​poor coating quality.

[0030] Each updated small area contains at most three coating difference factors of different sizes, namely 、 and ,if The size is equal to or , then there are only two coating difference factors of different sizes in each small area.

[0031] Traverse all small areas on the entire surface coating image, update the coating difference factor according to the above steps, and calculate the change in each area. Here, a coating poor quality feature is constructed to characterize local changes in the surface coating image. Local changes in the surface coating image are manifested as obvious edge changes and unstable gradient changes between the poor quality spray area and the background area. Calculate the coating poor quality feature I of each small area of ​​the entire surface coating image after the update:

[0032]

[0033]

[0034]

[0035] Where, is the coating quality feature of any region v in the surface coating image, are the sum of the coating difference factors in the first, second and third columns of the updated coating parameter matrix, is the local information difference of the image in the vertical direction within the small area after updating, are the sum of the coating difference factors in the first, second and third rows of the updated coating parameter matrix, It is the difference of local image information in the horizontal direction within the small area after update. is the amount of change in this area.

[0036] The coating inferior feature I reflects the difference in image information in the horizontal and vertical directions within a small area, and also reflects the difference in spraying quality within the small area. The larger the coating inferior feature I is, the more likely this small area contains an inferior coating area.

[0037] Traverse the entire surface coating image to obtain the coating inferior feature I corresponding to each small area. A total of m*n coating inferior features I are obtained for the surface coating image. All small areas are mapped one by one to each pixel to obtain the feature map F corresponding to the surface coating image. The coating inferior feature I value is mapped to the grayscale value of the pixel. The range of the mapped value is [0, .255]. The effective features in the feature map F are very sparse. Consider using a more compact feature to represent the changes in the pixel points in the feature map F. Use a 5*5 window to traverse the entire feature map F without repetition. For each traversed window area, obtain the gradient amplitude and gradient direction of the pixel point. Further, calculate the gradient histogram of the pixels in the area area, divide it into 20° intervals, and obtain the gradient histogram. Calculate the weight value of the corresponding angle according to the amplitude value in each angle interval. Calculating the weight value of the image gradient histogram is a well-known technology, and the specific calculation process will not be described in detail.

[0038] Observe the weights of all angles in the histogram. If each angle has a certain weight value, it is recorded as , , Always If the distribution between these weight values ​​is discrete, and the weight values ​​in different angle intervals have large differences, it means that the gradient changes of pixels in this area are relatively frequent and unequal, and there are more corresponding regional edges. Calculate the variance between the weight values ​​in all directions. If the variance is greater than the threshold T, the empirical value of the threshold T is 50, and it is considered that the pixels in this area vary drastically, which is reflected in the surface coating image. This means that there are obvious differences between the poor-quality coating features I in the window, and this window area corresponds to the poor-quality coating area.

[0039] Step 3: Combine the coating area results obtained in the above steps with different labels as the input of the detection model, and evaluate the quality of furniture spraying based on the output results of the network model.

[0040] Following the above steps, regions of the surface coating image containing inferior coating are identified. Minimum bounding rectangles (MBRs) of these regions are then obtained, allowing the locations of the inferior coating to be located within the surface coating image. Different labels are assigned to the MBRs, and the final finish quality of the furniture is inspected using a classification network. The size of the MBRs may vary, requiring normalization before input into the neural network. Image normalization is a well-known technique, and the detailed process is omitted here.

[0041] The labels of the neural network are annotated with the help of the Labelme tool. Label 0 corresponds to the sagging phenomenon in inferior coatings, label 1 corresponds to the bottom leakage phenomenon in inferior coatings, label 2 corresponds to the bubble phenomenon in inferior coatings, and label 3 corresponds to the dirty spot phenomenon in inferior coatings. The network structure adopts ResNet, with cross entropy as the loss function and Adam algorithm as the network optimization algorithm. The output of the neural network corresponds to different phenomena of inferior coating in the surface coating image. The spraying quality of the furniture is evaluated based on whether inferior spraying occurs after the furniture is sprayed and how many types of inferior spraying phenomena occur.

[0042] The larger the area of ​​inferior coating on the surface of a building material and the more types of inferior phenomena appear, the inferior quality of the spray coating of the building material furniture. Such building materials need to be removed from the assembly line and the furniture needs to be sprayed again. If there is no inferior coating area on the surface of the building material, it means that the quality of the spray coating of the building material furniture is high-quality.

Claims

1. A furniture spray quality detection method, characterized in that The following steps are involved: Step 1: Use an industrial camera to obtain images of the surface coating of furniture on building materials and pre-process the images; Step 2: Extract features from the pre-processed collected image and obtain the corresponding building material surface feature map based on the extracted features; Step 3: Combine the coating area results obtained in the above steps with different labels as the input of the detection model, and evaluate the quality of furniture spraying based on the output results of the network model.

2. The furniture spraying quality detection method according to claim 1, characterized in that The step one specifically includes: placing a CMOS camera above the processing line of the spray furniture process to collect surface images of building materials, performing noise reduction processing on the collected surface images of building materials, and installing a ring light source under the camera to obtain a pre-processed surface coating image.

3. The furniture spraying quality detection method according to claim 1 or 2, characterized in that The second step is specifically as follows: after acquiring the surface coating image formed by spraying the furniture, the local changes in the surface coating image are first obtained, and a certain number of grid forms are used as comparison objects when obtaining local features, thereby increasing image information and eliminating the influence of grid edgeization.

4. The furniture spraying quality detection method according to claim 1, characterized in that The step three is specifically as follows: according to the above steps, the area containing the inferior coating in the surface coating image is obtained, the minimum bounding rectangle of these areas containing the inferior coating is obtained, the position of the inferior coating is located in the surface coating image, different labels are assigned to the minimum bounding rectangle, and the final spraying quality of the furniture is detected through the classification network. The size of the minimum bounding matrix may be different, and size normalization processing is required before inputting into the neural network. The label of the neural network is annotated with the help of the Labelme tool. Label 0 corresponds to the sagging phenomenon in the inferior coating, label 1 corresponds to the bottom leakage phenomenon in the inferior coating, and label 2 corresponds to the bottom leakage phenomenon in the inferior coating. Label 1 corresponds to the bubble phenomenon in inferior coating, label 3 corresponds to the dirty spot phenomenon in inferior coating, and the output of the neural network corresponds to different phenomena of inferior coating in the surface coating image. The spraying quality of the furniture is evaluated based on whether inferior spraying occurs after the furniture is sprayed and how many types of inferior spraying phenomena occur. The larger the area of ​​inferior coating area on the surface of a building material and the more types of inferior phenomena appear, the inferior quality of the spraying of this building material furniture. Such building materials need to be removed from the assembly line for re-spraying of the furniture. If there is no inferior coating area on the surface of the building material, it means that the quality of the spraying of the building material furniture is high-quality.

5. The furniture spraying quality detection method according to claim 3, characterized in that The step 2 is specifically as follows: assuming that the size of the pre-processed surface coating image is M*N, the surface coating image is divided into m small areas of the same size, where the empirical value of m is 200, and any small area is divided into a*a small grids, where a is an empirical value of 5. At this time, each small area is divided into 5*5. In order to extract as much local image information as possible without generating information redundancy, a sliding window is set with a window size of 3*3. The window slides in the small area with a step size of 1 small grid without repetition. Each sliding result is regarded as a coating image component. Traversing the entire small area will obtain multiple coating image components, and the coating image components are non-repetitive. The coating image composition contains part of the image information. The similarities and differences between the image information contained in the pixels within the image composition reflect whether the current area contains different spraying effects. A coating difference factor is constructed here. The coating difference factor is used to characterize the grayscale difference between the pixels within a coating image composition. By comparing the coating difference factors of the central coating image composition and the surrounding adjacent coating image compositions in each 5*5 small area, the local change features that can characterize each small area are extracted from the comparison results.

6. The furniture spraying quality inspection method according to claim 5, further characterized by: Calculate the multidimensional features of each coating factor, traverse all small areas on the entire surface coating image, update the coating difference factor according to the above steps, and calculate the change amount of each area. Here, construct a coating inferior feature to characterize local changes in the surface coating image. Local changes in the surface coating image are manifested as obvious edge changes and unstable gradient changes between the inferior spray area and the background area; the coating inferior feature I reflects the difference in image information in the horizontal and vertical directions of the small area, and also reflects the difference between the quality of spraying in the small area. The larger the coating inferior feature, the more likely this small area contains an inferior coating area; Traverse the entire surface coating image to obtain the coating inferior feature I corresponding to each small area. A total of m*n coating inferior features I are obtained in the surface coating image. All small areas are mapped one by one to pixels to obtain the feature map F corresponding to the surface coating image. The coating inferior feature I value is mapped to the grayscale value of the pixel. The range of the mapped value is [0, .255]. The effective features in the feature map F are very sparse. Consider using a more compact feature to represent the changes in the pixel points in the feature map F. Use a 5*5 window to traverse the entire feature map F without repetition. For each traversed window area, obtain the gradient amplitude and gradient direction of the pixel point. Further, calculate the gradient histogram of the pixels in the area, with a 20° interval as the division interval, and obtain the gradient histogram.

Images