One-piece closestool surface defect detection method

Abstract

The invention discloses an integrated closestool surface defect detection method, and belongs to the technical field of integrated closestool defect detection. Comprising the following steps: carrying out video shooting on a to-be-detected one-piece closestool by adopting a combination of a posture correcting process and a posture reversing process, and carrying out frame extraction processing on the shot video to obtain a picture; performing structural similarity calculation on the picture subjected to frame extraction processing and each classification template picture, and performing matching to obtain each orientation classification picture of the to-be-detected one-piece closestool; and performing classification defect detection according to the matched classification pictures in all orientations. According to the method, toilet images can be obtained in all directions through normal posture and reverse posture shooting, the images are stored by adopting a frame extraction technology, classification processing is performed according to different surfaces of the toilet, the image processing time is greatly shortened, defect identification is realized by adopting a deep learning model and a related detection algorithm, the overall detection efficiency is remarkably improved, and the detection accuracy is improved. And the high-efficiency requirement of factory flow line production is met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of integrated toilet defect detection, and particularly relates to an integrated toilet surface defect detection method. BACKGROUND

[0002] With the improvement of people's living standards, the demand for home environment is increasing, and the quality of integrated toilet, as an important part of the bathroom, is directly related to the comfort and health of the user. Integrated toilet not only needs to have good functionality, but also needs to be flawless in appearance to meet consumers' pursuit of high-quality life. However, in the production process of integrated toilet, due to the complexity and diversity of manufacturing process, various defects often occur. These defects not only affect the appearance of the product, but also may affect its service life and functionality, and even lead to consumer complaints or product recall. Therefore, how to effectively detect and eliminate these defects has become the focus of related enterprises.

[0003] Traditional integrated toilet defect detection mainly relies on manual visual inspection. Workers on the production line identify cracks, color differences, uneven glaze and other problems on the surface of the product by visual inspection. However, manual visual inspection has significant limitations. First, the accuracy of manual detection depends on the experience and skills of workers, and these factors vary greatly among different workers. This subjectivity leads to poor repeatability of detection results, and different workers may make different judgments on the same product defects. Second, long-term repetitive work can easily lead to worker fatigue, further reducing the accuracy of detection. With the improvement of production speed, workers need to complete the detection in a shorter time, which increases the probability of missed detection and false detection. In addition, manual detection is relatively low in efficiency, which is difficult to meet the needs of modern large-scale production.

[0004] With the development of technology, the level of industrial automation is constantly improving, and automated defect detection methods based on image processing and computer vision technology have been gradually applied to the quality control of various industrial products. These methods capture the image of the product surface through the camera, use algorithms for image processing and analysis, and identify possible defects. Compared with manual detection, this method has the advantages of fast speed, high precision and good stability, and can achieve efficient quality detection in large-scale production.

[0005] However, traditional image processing methods still have deficiencies when facing complex surface defects of a one-piece toilet. The surface structure of a one-piece toilet is complex, with various shapes, materials, and color changes. Traditional image processing algorithms often perform poorly when dealing with these complexities. In addition, the types and forms of defects are diverse, and some defects have subtle shapes and unobvious color changes, making it difficult for traditional algorithms to accurately identify them. To improve the accuracy and efficiency of defect detection, a more intelligent and robust detection method is urgently needed. SUMMARY

[0006] To overcome the deficiencies of the prior art, the invention uses front and back posture shooting to obtain images of the toilet from all directions, uses frame extraction technology to save images, and classifies and processes different surfaces of the toilet, significantly shortening image processing time. At the same time, the invention uses a deep learning model and related detection algorithms to identify defects, significantly improving overall detection efficiency.

[0007] To achieve the above purpose, the invention provides a one-piece toilet surface defect detection method, comprising the following steps:

[0008] (1) Video shooting of the one-piece toilet to be detected is performed using two processes of front and back posture, and frame extraction processing is performed on the shooting video to obtain pictures;

[0009] (2) Structure similarity calculation is performed on the pictures after frame extraction processing and each classification template picture, and each orientation classification picture of the one-piece toilet to be detected is matched;

[0010] The front posture classification pictures include front view, left view, right view, rear view, left downward view, and right downward view. The back posture classification pictures include front view, left view, right view, rear view, left upward view, front upward view, and right upward view.

[0011] (3) Classification defect detection is performed according to each orientation classification picture matched;

[0012] The front view, left view, and right view pictures of the one-piece toilet in front posture are subjected to surface defect detection using a trained deep learning model;

[0013] The left downward view and right downward view pictures of the one-piece toilet in front posture are subjected to straight line fitting method to determine the curvature of the toilet edge;

[0014] The rear view picture of the one-piece toilet in front posture is also subjected to symmetry deformation inspection using similar structure calculation;

[0015] All pictures of the one-piece toilet in back posture are subjected to surface defect detection using a trained deep learning model;

[0016] For the obtained front view of the connected toilet upside down, the toilet drain area ROI is extracted, edge extraction and Hough circle detection are used to extract the drain area features for detecting enamel defects.

[0017] Further, the upright position is the connected toilet upright, and the upside-down position is the connected toilet with the water tank placed back down.

[0018] Further, the connected toilet shooting adopts a device including a toilet rotating base, two cameras, and a standard light source.

[0019] When shooting in the upright position, the two cameras are placed at the same distance above and below the toilet rotating base, the lower camera is directly opposite the connected toilet, and can capture a complete image of the side of the connected toilet; the upper camera looks down, and can capture the side and edge profile of the connected toilet.

[0020] When shooting in the upside-down position, the two cameras are placed at different distances above and below the toilet rotating base, the upper camera is directly opposite the connected toilet, and can capture images of the inside of the toilet tank, the bottom water outlet, and the base; the lower camera looks up, and can capture images of the side water outlet.

[0021] Further, the classification template pictures are obtained by shooting the upright and upside-down positions of the selected perfect samples of the to-be-inspected toilet style.

[0022] Further, the step (2) is specifically:

[0023] (2.1) Obtain the classification template pictures of the connected toilet;

[0024] (2.2) Calculate the structural similarity between the saved pictures after frame extraction and the classification template pictures, and compare the similarity.

[0025]

[0026] Where: x and y are two images to be compared, μ x and μ y are the mean values of images x and y: and are the variances of images x and y, respectively: σ xy is the covariance between images x and y: C1 and C2 are constants.

[0027] (2.3) For each type of picture, select the picture with the highest similarity to the classification template picture as the picture of that type of connected toilet.

[0028] Further, the left overhead view and the right overhead view of the connected toilet in the normal posture are fitted with a straight line of the toilet edge using a random sample consensus algorithm for defect detection, specifically:

[0029] (3.1) A specific area is selected as the ROI of the edge of the connected toilet, and preprocessed through grayscale processing and histogram equalization;

[0030] (3.2) The edge line of the toilet in the ROI of the preprocessed connected toilet edge is detected, and all end points are extracted to form a point set;

[0031] (3.3) A minimum number of data points are randomly selected from the point set to fit a straight line;

[0032] (3.4) The perpendicular distance of all data points in the point set to the fitted straight line is calculated, and the points satisfying the threshold condition are the inliers, and the points not satisfying the threshold condition are the outliers;

[0033] (3.5) Steps (3.3)-(3.4) are repeated until a fitted straight line with the most inliers is found;

[0034] (3.6) The proportion of all inliers satisfying step (3.5) to all data points in the point set is calculated, and if the proportion is greater than a threshold, it is considered that the edge of the toilet has no defect, otherwise, it is considered that the edge of the toilet has a bending defect.

[0035] Further, the rear view of the connected toilet in the normal posture is also subjected to symmetry calculation and deformation judgment using similar structural calculation, specifically: after cutting and rotating the rear view of the connected toilet in the normal posture along the central axis, the structural similarity of the cut picture is calculated to obtain a similarity number, and if the similarity number is greater than a set threshold, it is considered that the rear view of the toilet satisfies the symmetry and has no deformation, otherwise, it is considered that there is a deformation defect.

[0036] The beneficial effects of the present application are:

[0037] The present application can obtain images of the toilet in all directions by shooting in the normal posture and the inverted posture, save the images using frame extraction technology, classify and process the images according to different surfaces of the toilet, greatly shorten the image processing time, and use a deep learning model and related detection algorithms for defect recognition. Therefore, the present application can automatically process from image acquisition to processing and defect recognition, reduce the requirement for the skills of the operator, improve the operation convenience and overall detection efficiency of the system, and meet the efficient needs of the factory assembly line production. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 It is a flowchart of the surface defect detection method of the connected toilet according to the embodiment of the present application.

[0039] Figure 2The figure shows a normal posture photographing schematic diagram of the one-piece toilet according to the embodiment of the present application.

[0040] Figure 3 The figure shows an upside-down posture photographing schematic diagram of the one-piece toilet according to the embodiment of the present application.

[0041] Figure 4 The figure shows a normal posture and upside-down posture classification schematic diagram of the one-piece toilet according to the embodiment of the present application.

[0042] Figure 5 The figure shows a normal posture classification schematic diagram of the one-piece toilet according to the embodiment of the present application.

[0043] Figure 6 The figure shows an upside-down posture classification schematic diagram of the one-piece toilet according to the embodiment of the present application.

[0044] Figure 7 The figure shows a side pinhole and collision defect schematic diagram of the toilet according to the embodiment of the present application.

[0045] Figure 8 The figure shows a base edge straight line detection schematic diagram of the toilet according to the embodiment of the present application.

[0046] Figure 9 The figure shows a rear view symmetry detection schematic diagram of the toilet according to the embodiment of the present application.

[0047] Figure 10 The figure shows an edge water outlet inner side defect detection schematic diagram of the toilet according to the embodiment of the present application.

[0048] Figure 11 The figure shows a sewage outlet green edge schematic diagram of the toilet according to the embodiment of the present application. DETAILED DESCRIPTION

[0049] The present application will be further described below in combination with the drawings and embodiments.

[0050] As shown in the figure, the present application provides a one-piece toilet surface defect detection method, which comprises the following steps: Figure 1

[0051] S101, video photographing is performed on the one-piece toilet to be detected by using two process combinations of normal posture and upside-down posture, and picture is obtained by frame extraction processing on the photographed video;

[0052] As shown in the figure, Figure 2 Figure 3 The one-piece toilet photographing is performed by using a device comprising a toilet rotating base, two cameras and a standard light source. According to the embodiment of the present application, video photographing is performed on the one-piece toilet by using two process combinations of normal posture and upside-down posture. The normal posture is that the one-piece toilet is placed normally, and the upside-down posture is that the water tank of the one-piece toilet is placed with the back downward.

[0053] ​​When shooting in the correct position, the two cameras are placed one above the other at the same distance from the rotating base of the toilet. The lower camera faces the one-piece toilet and can capture a complete image of the side of the one-piece toilet; the upper camera looks down and can capture the side and edge outline of the one-piece toilet.

[0054] When shooting in an inverted position, the two cameras are placed up and down at different distances from the rotating base of the toilet. The upper camera faces the one-piece toilet and can capture images of the toilet tank, the bottom water outlet and the base; the lower camera looks up and can capture images of the side water outlet.

[0055] like Figure 2 、 Figure 3 As shown, in this embodiment, the toilet rotating base 1 is required to rotate one circle in 20 seconds, so its angular velocity ω is:

[0056]

[0057] When shooting in the upright position, Camera 3 is a 20-megapixel ultra-high-definition industrial camera, positioned 30cm above the ground and 150cm horizontally from the toilet, enabling clear side-on images of the toilet. Camera 4 is a 5-megapixel high-definition industrial camera, positioned 100cm above the ground and 150cm horizontally from the toilet. This downward-looking camera can clearly eliminate surface irregularities and edge deformation. Light source 5 is a standard light source, used for supplemental lighting when photographing the toilet. Placed directly behind the camera, the light reflects perpendicularly to the toilet surface, enabling the camera to capture surface defects such as small dents (orange peel glaze).

[0058] When shooting in the inverted position, Camera 3 is positioned the same as when shooting in the upright position, capturing images of the toilet tank, the bottom outlet, and the base. Camera 4 is positioned 10 cm above the ground and 90 cm horizontally from the toilet, looking upward to capture images of the side outlet.

[0059] This embodiment of the present invention achieves omnidirectional imaging by rotating the toilet, requiring only two fixed-position industrial cameras to complete the task, reducing hardware complexity and lowering costs. Furthermore, this solution is highly compatible with the requirements of factory automation production lines, facilitating its application in real-world production environments. To capture the image, the toilet is secured to a rotating table, which is then activated simultaneously with the detection program. A high-definition industrial camera records the rotating toilet, generating a video.

[0060] The captured video is saved every N frames by using the video frame extraction method to reduce the amount of data for subsequent processing.

[0061] S102, structure similarity calculation is performed on the picture after the frame extraction processing and each classification template picture, and each orientation classification picture of the to-be-detected integrated toilet is matched.

[0062] As shown in Figure 4 , the upright posture classification of the integrated toilet includes front view, left view, right view, rear view, left downward view and right downward view; and the upside-down posture classification of the integrated toilet includes front view, left view, right view, rear view, left upward view, front upward view and right upward view.

[0063] The unflawed front sample of the to-be-inspected toilet style is selected for upright posture and upside-down posture shooting, and each classification template picture is obtained according to the above classification as a template picture.

[0064] The picture after the frame extraction is saved, and the template picture of each classification is calculated by using structure similarity, and similarity comparison is performed.

[0065] SSIM is an index for measuring the similarity of two images, which is the product of three components of brightness l, contrast c and structure s.

[0066] SSIM(x,y)=[l(x,y)] α ·[c(x,y)] β ·[s(x,y)] γ

[0067] l is used to measure the average brightness difference of the image, and the formula is as follows:

[0068]

[0069] C is used for contrast comparison, and the formula is as follows:

[0070]

[0071] S reflects the similarity of local structure of the image, and the formula is as follows:

[0072]

[0073] Therefore:

[0074]

[0075] Wherein: x and y are two images to be compared, μ x and μ y are the mean values of the images x and y: and are the variances of the images x and y: σ xy is the covariance between the images x and y: C1 and C2 are constants, which are used to prevent the denominator from being zero.

[0076] C1=(K1L) 2 ,C2=(K2L) 2

[0077] Wherein: L is the dynamic range of image pixel value, for example, L of 8 is 255 for gray scale image. K1 and K2 are very small constants, generally K1, K2 are 0.01 and 0.03.

[0078] For each type of picture, the picture with the highest similarity to the template picture is selected as the picture of the connected toilet of this classification.

[0079] Through this operation, the standard sampling of pictures of all surfaces of the connected toilet can be quickly completed, while reducing the sample size, and also facilitating defect positioning of the detection algorithm and individual customization of the detection process of each surface.

[0080] S103, according to the classified pictures of each orientation obtained by matching, performing classification defect detection.

[0081] In the embodiment of the application, the deep learning model selects a YOLOv8 detection model for defect detection. The model is obtained after training using a large amount of defect data labeled by artificial labeling.

[0082] The video pictures of the normal posture sampling will respectively retain the front view, left view, right view, rear view, left overhead view and right overhead view of the toilet, and are respectively denoted as A1, A2, A3, A4, A5 and A6. The six pictures are respectively detected, as shown in the following table. Figure 5

[0083] For the four pictures of A1, A2, A3 and A4, the YOLOv8 detection model trained by a large amount of defect data is used to detect defects such as pinholes and scratches, and the detection time of each picture is about 150ms, and the detection time of four pictures is 600ms. For example, as shown in the following table. Figure 7 The detected side pinhole defect.

[0084] For the two pictures of A5 and A6, since they are matched from the template pictures, a specific area can be selected as the ROI of the edge of the toilet, and the RANSAC (Random Sample Consensus) algorithm is used to fit the straight line of the edge of the toilet, so as to judge the curvature of the straight line segment, as follows:

[0085] (1) A specific area is selected as the ROI of the edge of the connected toilet, and is preprocessed by gray processing and histogram equalization.

[0086] (2) The edge line of the connected toilet in the preprocessed ROI of the edge of the connected toilet is detected, and all end points are extracted to form a point set.

[0087] ​(3) Randomly select the minimum number of data points from the point set to fit a straight line.

[0088] The linear equation is selected for fitting in the embodiment of the present application:

[0089] y = ax + b,

[0090] wherein a is the slope and b is the intercept.

[0091] (4) Calculate the perpendicular distance of all data points in the point set to the fitted straight line, and the points whose perpendicular distance meet the threshold condition are the inliers, and the points whose perpendicular distance do not meet the threshold condition are the outliers.

[0092] This is a consistency detection, and the perpendicular distance formula is as follows:

[0093]

[0094] (5) Repeat steps (3)-(4) until a fitted straight line with the most inliers is found.

[0095] (6) Calculate the proportion of all inliers meeting step (5) to all data points in the point set, and if the proportion is greater than the threshold value, it is considered that the one-piece toilet has no edge defect, otherwise, it is considered that the one-piece toilet has an edge defect.

[0096] In the embodiment of the present application, the threshold value is set to 0.8, and if the proportion of inliers is greater than 0.8, it indicates that most points fall near the fitted straight line, and the line segment is considered to be "straight"; otherwise, if the proportion of inliers is low, it indicates that the line segment is "curved or mixed".

[0097] The detection result is shown in Figure 8 .

[0098] For the A4 picture, a similar structural calculation is also needed to verify the symmetry and make a deformation judgment. The A4 picture is cut along the central axis and rotated, and then the structural similarity comparison of the cut picture is performed using the structure similarity comparison in step S101(3), to obtain the similarity degree of the two pictures. Through setting a threshold value, the similarity degree greater than the given threshold value A can be considered to satisfy the symmetry, i.e., no deformation, otherwise, it is considered to have deformation. Here, the threshold value A is set to 0.8, and the detection result is shown in Figure 9 .

[0099] The video pictures of the upside-down sampling will respectively retain the front view, left view, right view, rear view, left overhead view, front overhead view, right overhead view, and seven pictures, respectively recorded as B1, B2, B3, B4, B5, B6, and B7, for detection, as shown in Figure 6 .

[0100] The trained YOLOv8 model consistent with the same normal posture detection is used for detection of the above-mentioned 7 pictures, and the postures can be used to detect surface defects appearing in the toilet tank, the base, the side of the toilet, and the inside of the edge water outlet, respectively. The time consumption of single picture detection is about 150 ms, and the time consumption of four pictures detection is 600 ms. The surface defects inside the edge water outlet are as shown in Figure 10

[0101] For B1, a specific interest region ROI is extracted to obtain a toilet drain area graph. Due to the complex shape, the edge will appear glaze defects, resulting in a blue color. The region is located, edge extraction and Hough circle detection are used to extract the drain area features for detecting glaze defects, as shown in Figure 11

[0102] After the detection is completed, the defect pattern and the detection result are saved, and finally a report form is displayed to the user, which is convenient for the user to locate the defect types existing in the production line and help to improve the specific links in the production line.

[0103] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the principles and spirit of the present application shall be included in the protection scope of the present application.​​

Claims

1. A method for detecting surface defects of a one-piece toilet, characterized in that: The steps include: (1) A video of the one-piece toilet to be inspected is shot using a combination of two processes: the upright position and the inverted position, and the shot video is processed by frame extraction to obtain an image; (2) performing structural similarity calculation on the image after the frame extraction process and each classification template image, and matching them to obtain classification images of each direction of the one-piece toilet to be detected; The upright posture classification pictures include front view, left view, right view, rear view, left top view, and right top view; the inverted posture classification pictures include front view, left view, right view, rear view, left top view, front top view, and right top view; (3) Classification defect detection is performed based on the matched classification images of various orientations; Using the trained deep learning model, surface defect detection is performed on the obtained front view, left view, and right view images of the one-piece toilet in the correct posture. Using a straight line fitting method on the left and right top-down pictures of the one-piece toilet in the correct posture, the straightness of the toilet edge is determined; The rear view of the one-piece toilet in the upright position is also subjected to symmetric deformation inspection using similar structural calculation; All the images of the one-piece toilet in an inverted position are subjected to surface defect detection using a trained deep learning model; The toilet drain outlet region ROI is extracted from the obtained front view of the one-piece toilet in an inverted position, and edge extraction and Hough circle detection are used to extract drain outlet region features for detecting glaze defects.

2. The method for detecting surface defects of a one-piece toilet according to claim 1, characterized in that: The upright posture is when the one-piece toilet is placed upright, and the inverted posture is when the one-piece toilet water tank is placed with its back facing downwards.

3. The method for detecting surface defects of a one-piece toilet according to claim 1, characterized in that: The one-piece toilet is photographed using a device including a toilet rotating base, two cameras, and a standard light source; When shooting in the upright position, the two cameras are placed at the same distance from the toilet rotating base. The lower camera faces the one-piece toilet and can capture a complete side image of the one-piece toilet; the upper camera looks down and can capture the side and edge outline of the one-piece toilet. When shooting in an inverted position, the two cameras are placed up and down at different distances from the rotating base of the toilet. The upper camera faces the one-piece toilet and can capture images of the toilet tank, the bottom water outlet and the base; the lower camera looks up and can capture images of the side water outlet.

4. The method for detecting surface defects of a one-piece toilet according to claim 1, characterized in that: The template images for each classification are obtained by selecting flawless positive samples of the toilet style to be inspected and photographing them in the upright and inverted positions.

5. The method for detecting surface defects of one-piece toilets according to claim 1, characterized in that: The step (2) is specifically as follows: (2.1) Obtain template images of each category of one-piece toilets; (2.2) Comparing the saved image with the classification template images using structural similarity calculation; Where: x and y are two images to be compared, μ x and μ y is the mean of the images x and y: and are the variances of images x and y respectively: σ xy is the covariance between images x and y: C1 and C2 are constants; (2.3) For each category of pictures, the picture with the highest similarity to each category template picture is selected and retained as the picture of the one-piece toilet in that category.

6. The method for detecting surface defects of one-piece toilets according to claim 1, characterized in that: The defect detection using the random sampling consistency algorithm to fit the toilet edge straight line on the left and right top view pictures of the one-piece toilet is specifically as follows: (3.1) Selecting a specific area as the ROI of the edge of the one-piece toilet and performing grayscale processing and histogram equalization preprocessing; (3.2) detecting the toilet edge line segments within the ROI of the one-piece toilet edge after preprocessing, and extracting all endpoints to form a point set; (3.3) randomly selecting a minimum number of data points from the set of points to fit a straight line; (3.4) Calculate the vertical distances from all data points in the point set to the fitted line. Points whose vertical distances meet the first threshold condition are considered inliers, and those that do not meet the threshold condition are considered outliers. (3.5) Repeat steps (3.3)-(3.4) until the fitting line with the most inliers is found; (3.6) Calculate the ratio of all internal points that meet step (3.5) to all data points in the point set. If this ratio is greater than the second threshold, it is considered that the toilet edge has no defects. Otherwise, it is considered that the toilet edge has a bending defect.

7. The method for detecting surface defects of a one-piece toilet according to claim 1, characterized in that: The rear view of the one-piece toilet in the correct posture is also checked for symmetry and deformation judgment using similar structural calculation. Specifically, the rear view of the one-piece toilet in the correct posture is cut along the central axis and rotated, and the structural similarity calculation is performed on the cut image to obtain the similarity degree. If the similarity degree is greater than the third threshold, it can be considered that the rear view of the toilet satisfies the symmetry, that is, there is no deformation; otherwise, it is considered that there is a deformation defect.

Images