A method for detecting punching defects of the inner liner plate of a washing machine

Abstract

The present invention relates to the technical field of image processing, and particularly relates to a method for detecting punching defects of an inner tank plate of a washing machine. The present invention obtains an approximate circular region by acquiring a grayscale image of the punching of the inner tank plate, screens out regions with a higher circularity in the punching image of the inner tank plate, analyzes the features within the approximate circular region, obtains an approximate circular region that conforms to the features inside the punching within the approximate circular region, and analyzes the features of high specular interference around the approximate circular region, constructs a halo feature coefficient and a light spot feature coefficient, obtains an approximate circular region with a high specular feature that is more in line with the punching edge, thereby obtaining a region with punching features, selects the region with punching features for marking, and then uses the watershed algorithm to segment the grayscale image of the punching of the inner tank plate, avoids the problem of over-segmentation of the watershed algorithm, obtains a more accurate punching region, and improves the detection accuracy of punching defects of the inner tank plate of the washing machine.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a method for detecting punching defects of a washing machine inner liner plate. Background Art

[0002] As an indispensable household appliance in modern families, the stability of the performance and the length of the service life of a washing machine are directly related to the user experience. The inner liner plate of a washing machine, as one of the core structural components of the washing machine, not only bears various physical effects during the washing process of clothes, but also is responsible for the drainage function of the washing machine. Therefore, the punching quality of the inner liner plate has a crucial impact on the drainage performance, structural strength and overall service life of the washing machine.

[0003] The inner liner plate is usually made of stainless steel material because it has good corrosion resistance and sufficient mechanical strength to meet the requirements of the washing machine during long-term use. During the production process, the inner liner plate forms regularly arranged drainage holes or functional holes through the stamping process. The design of these holes not only concerns the drainage efficiency of the washing machine, but also affects the water flow distribution and washing effect inside the washing machine.

[0004] However, during the punching process, due to the influence of various factors, such as the wear of the mold, the uneven distribution of material stress, the fluctuation of equipment accuracy, etc., it is often difficult to fully guarantee the punching quality of the inner liner plate. These factors are likely to cause various defects in the inner liner plate, such as hole position deviation, hole diameter out-of-tolerance (i.e., the hole diameter is greater than or less than the design value), edge burrs, micro-cracks and deformation, etc.

[0005] The influence of these punching defects on the performance and service life of the washing machine is significant. Hole position deviation and hole diameter out-of-tolerance may damage the sealing performance of the inner liner plate, resulting in water leakage during the operation of the washing machine; edge burrs and micro-cracks may weaken the mechanical properties of the inner liner plate, making it a weak link in the washing machine structure and prone to breakage or damage during long-term use; and deformation may change the shape and size of the inner liner plate, affecting its fitting accuracy with other components of the washing machine, and further causing problems such as abnormal noise or structural failure.

[0006] Therefore, punching defect detection has become a key link in the quality control of washing machine production. Traditional defect detection methods, such as visual inspection or using ordinary optical detection equipment, are often easily interfered by the high reflective characteristics of the stainless steel surface, resulting in a reduction in the accuracy and efficiency of defect detection. Especially when detecting subtle defects such as edge burrs and micro-cracks, the high reflective surface may make it difficult to accurately identify these defects, thus affecting the overall quality control of the washing machine.

[0007] In summary, in the prior art, there is a problem of insufficient accuracy in detecting punching defects of the inner tank plate of a washing machine. There is an urgent need for a new technology or method that can overcome the high specular reflection interference on the stainless-steel surface and improve the accuracy and efficiency of defect detection. Summary of the Invention

[0008] The purpose of the present invention is to provide a method for detecting punching defects of the inner tank plate of a washing machine, so as to solve the problem that the existing defect detection method is easily interfered by the high specular reflection on the stainless-steel surface and affects the accuracy of defect detection. To this end, the present invention provides the following technical solutions to solve the above problems.

[0009] The present invention provides a method for detecting punching defects of the inner tank plate of a washing machine, including:

[0010] Obtain the closed edge line of the gray-scale image of the inner tank plate punching to obtain an approximate circular area in the area enclosed by the closed edge line;

[0011] Divide the approximate circular area into a suspected light spot area and a light ring area;

[0012] Calculate the light spot characteristic coefficient of each suspected light spot area, and use the suspected light spot area with the light spot characteristic coefficient greater than or equal to the light spot characteristic threshold as the light spot area; the light spot characteristic coefficient is positively correlated with the contrast and the light spot consistency coefficient, and the light spot consistency coefficient characterizes the consistency of the light spot appearance positions in each suspected light spot area;

[0013] Use either the light ring area or the light spot area as the target area; calculate the hole-in characteristic coefficient of the target area, and the hole-in characteristic coefficient is inversely correlated with the variance and mean value of the gray-scale values of all pixel points in the target area;

[0014] Use the target area with the hole-in characteristic coefficient greater than the hole-in characteristic threshold as the punching characteristic area;

[0015] Centering on each punching characteristic area, segment the gray-scale image of the inner tank plate punching, and perform defect detection on the segmented gray-scale image to obtain punching defects.

[0016] The above solution analyzes the characteristics of the high - specular interference of the punched holes in the gray - scale image of the inner liner plate by obtaining the gray - scale image of the punched holes in the inner liner plate. First, an approximate circular area is obtained, and the areas with higher circularity in the punched - hole image of the inner liner plate are screened out. Then, the characteristics within the approximate circular area are analyzed to obtain an approximate circular area that conforms to the characteristics inside the punched hole, and the characteristics of the high - specular interference around the approximate circular area are analyzed. The halo feature coefficient and the light - spot feature coefficient are constructed to obtain an approximate circular area with high - specular characteristics that better conforms to the punched - hole edge, thereby obtaining the area with punched - hole characteristics. The area with punched - hole characteristics is selected for marking, and then the watershed algorithm is used to segment the gray - scale image of the punched holes in the inner liner plate, avoiding the problem of over - segmentation of the watershed algorithm, obtaining a more accurate punched - hole area, and improving the detection accuracy of the punched - hole defects of the washing - machine inner liner plate.

[0017] Optionally, the method of dividing the approximate circular area into a suspected light - spot area and a halo area includes: taking the approximate circular area with a halo feature coefficient less than the halo feature threshold as the suspected light - spot area, and taking the approximate circular area with a halo feature coefficient greater than or equal to the halo feature threshold as the halo area;

[0018] The halo feature coefficient hfc is: ; where amp min is the mean of the gradient magnitudes of all pixel points on the minimum circumscribed circle of each approximate circular area, amp cir is the mean of the gradient magnitudes of all pixel points on the peripheral circle adjacent to the minimum circumscribed circle of each approximate circular area, var gra is the variance of the gray - scale values of all pixel points on all peripheral circles of each approximate circular area, k min is the absolute value of the slope of the minimum fitting line of each approximate circular area, k cir is the absolute value of the slope of the minimum adjacent fitting line of each approximate circular area; is a preset tuning coefficient, and sigmoid() is a normalization function.

[0019] The above solution calculates the halo feature coefficient of each approximate circular area through the average situation, dispersion degree of the gray - scale values of the pixel points on the minimum circumscribed circle and peripheral circles of the approximate circular area, and the slopes of the minimum fitting line and the minimum adjacent fitting line, thereby improving the accuracy of halo feature recognition.

[0020] Optionally, the process of obtaining the minimum fitting line is specifically as follows:

[0021] For each approximate circular area in the gray - scale image of the inner liner plate, the HOG algorithm is used to obtain the gradient - direction histogram of all pixel points on the minimum circumscribed circle of the approximate circular area;

[0022] The values of all intervals in the gradient - direction histogram are subjected to linear fitting processing to obtain the minimum fitting line.

[0023] Optionally, the in-hole feature coefficient is specifically calculated as follows:

[0024] ;

[0025] where is the in-hole feature coefficient of the target area; var cir is the variance of the gray values of all pixel points in the target area, ave is the average of the gray values of all pixel points in the target area, and ave def is the preset gray average value; is the preset tuning parameter factor.

[0026] The above solution effectively improves the detection accuracy and efficiency by obtaining the edge line in the gray image of the inner liner plate punching, calculating the circularity of the area enclosed by the closed edge line, and taking the area enclosed by the closed edge line with a circularity greater than the preset circularity threshold as the approximate circular area, and obtaining the in-hole feature coefficient of the approximate circular area, and optimizing the quality control of the inner liner plate punching, thereby improving the product reliability and safety.

[0027] Optionally, the approximate circular area is the area enclosed by the closed edge line with a circularity greater than the preset circularity threshold.

[0028] Optionally, the spot feature coefficient is specifically obtained as follows:

[0029] where ffc is the spot feature coefficient of each suspected spot area; ind fac is the spot consistency coefficient of the clustering cluster where the minimum circumscribed circle of each suspected spot area is located, and ldc is the light and dark contrast of the minimum circumscribed circle of each suspected spot area; sigmoid() is the normalization function.

[0030] The above solution obtains the spot feature coefficient of the minimum circumscribed circle of each approximate circular area through the spot consistency coefficient and the light and dark contrast of the minimum circumscribed circle of each suspected spot area, thereby improving the accuracy of spot recognition.

[0031] Optionally, the spot consistency coefficient is specifically:

[0032] ;

[0033] where ind fac is the spot consistency coefficient of the clustering cluster; average() is the average value function, It is a set composed of the silhouette coefficients of all feature points in the clustering cluster. The clustering cluster is obtained by taking the feature vectors of the minimum circumscribed circle of the suspected light spot area as feature points and performing clustering processing on all feature points in the neighborhood of the minimum circumscribed circle of the suspected light spot area. The feature vectors are obtained by processing the pixel points in the minimum circumscribed circle of the suspected light spot area using the SIFT algorithm.

[0034] Optionally, the process of obtaining the defects of the inner liner plate punching specifically includes:

[0035] For the grayscale image of the inner liner plate punching, use the marker-based watershed algorithm to segment it. Take the area with a radius of a preset radius centered on each punching feature area as the marker area, where the value of the preset radius should be less than the radius of the punching.

[0036] Perform image segmentation on the grayscale image of the inner liner plate punching within the marker area to obtain a segmented image. Use the template matching algorithm based on the image pyramid to detect defects on the segmented grayscale image of the inner liner plate punching to obtain the punching defects of the inner liner plate.

[0037] The above solution performs image segmentation on the grayscale image of the inner liner plate punching within the marker area to obtain a segmented image, uses the template matching algorithm based on the image pyramid to detect defects on the segmented grayscale image of the inner liner plate punching, and obtains the punching defects of the inner liner plate, thereby improving the accuracy and efficiency of defect detection.

[0038] Optionally, the method for marking the grayscale image of the inner liner plate punching includes: using a marking algorithm to mark the position of the punching feature area on the image.

[0039] Optionally, the grayscale image of the inner liner plate punching is obtained by using a high-pixel industrial camera to collect the inner liner plate punching image, performing denoising processing on the inner liner plate punching image using image denoising, and then performing grayscale processing on the denoised surface image of the inner liner plate punching, where the image denoising is performed using the mean filter denoising algorithm.

[0040] The beneficial effects of the present invention are:

[0041] The solution of the present invention can analyze the characteristics of high - reflective interference of the punching holes on the inner liner plate punching grayscale image by obtaining the inner liner plate punching grayscale image. First, an approximate circular area is obtained, and the areas with higher circularity in the inner liner plate punching image are screened out. Then, the characteristics within the approximate circular area are analyzed to obtain an approximate circular area that conforms to the characteristics inside the punching hole, and the characteristics of high - reflective interference around the approximate circular area are analyzed to construct a halo feature coefficient and a light spot feature coefficient, so as to obtain an approximate circular area with high - reflective characteristics that is more in line with the punching edge, thereby obtaining the area with punching characteristics. Select the area with punching characteristics for marking, and then use the watershed algorithm to segment the inner liner plate punching grayscale image to avoid the problem of over - segmentation of the watershed algorithm, obtain a more accurate punching area, and improve the detection accuracy of the punching defects of the washing machine inner liner plate. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 FIG. schematically shows a flowchart of the steps of a method for detecting punching defects of a washing machine inner liner plate in this embodiment;

[0043] Figure 2 FIG. schematically shows a schematic diagram of the outer circle in a method for detecting punching defects of a washing machine inner liner plate in this embodiment;

[0044] Figure 3 FIG. schematically shows a histogram of gradient directions in a method for detecting punching defects of a washing machine inner liner plate in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts fall within the protection scope of the present invention.

[0046] The present invention aims at a solution for detecting punching defects of a washing machine inner liner plate. The punching defect detection solution of the inner liner plate needs to obtain the punching feature area by using the inner - hole feature coefficient, the halo feature coefficient, and the light - spot feature coefficient, mark the inner liner plate punching grayscale image according to the punching feature area, and segment the inner liner plate punching grayscale image to complete the detection of the punching defects of the washing machine inner liner plate.

[0047] Therefore, based on the above content, the present invention provides a method for detecting punching defects of a washing machine inner liner plate to accurately identify the detection of punching defects on the washing machine inner liner plate in a high - reflective environment.

[0048] Specifically, as Figure 1As shown in the figure, a method for detecting punching defects of the inner tank plate of a washing machine in this embodiment includes the following steps:

[0049] Step S1: Collect the closed edge line of the grayscale image of the punching on the inner tank plate of the washing machine to obtain an approximate circular area in the area enclosed by the closed edge line.

[0050] Specifically, the present invention selects the inner tank plate of the washing machine made of steel for defect detection. During the production and processing of the inner tank plate made of steel of the washing machine, usually with the steel plate as a flat plate, regular distribution of drainage holes is punched on the steel plate, and then the punched flat plate is curled into a cylindrical shape, and the closed joint is welded. In order to obtain a more accurate image, after punching regular distribution of drainage holes on the inner tank plate and before curling it into a cylindrical shape and welding, a high-pixel industrial camera is placed on the front to collect the surface image of the punching on the inner tank plate, and image denoising is used to perform denoising processing on the punching image of the inner tank plate. The denoising algorithm selected in the embodiment of the present invention is the mean filtering denoising algorithm, and the implementer can select other image denoising algorithms according to the actual situation. The surface image of the punching on the inner tank plate after denoising is grayscale processed to obtain the grayscale image of the punching on the inner tank plate.

[0051] In this embodiment, the inner tank plate of the washing machine selected by the present invention is made of steel, and the inner tank plate made of steel is processed by stamping. During the process of obtaining the punching during stamping, the inner tank plate usually has the defect of missing hole shape.

[0052] Under normal circumstances, the punching on the inner tank plate should be circular. When the inner tank plate has the defect of missing hole shape, among them, the missing hole shape includes burrs, inconsistent hole diameters, hole position offsets, and shape deformations. The missing hole shape may affect the performance of the washing machine. The abnormal shape and distribution of the punching will hinder the rapid drainage of water, and the incomplete or uneven distribution of the hole shape of the punching will disrupt the normal circulation path of the water flow during washing, reducing the tumbling and friction efficiency of the clothes.

[0053] When detecting the defects of the punching of stamping parts based on traditional image processing technology, usually the punching part of the stamping part is obtained based on the image segmentation algorithm, and then the punching defects are detected; however, when using the watershed algorithm to segment the punching defects of the inner tank plate, due to the presence of reflective areas in the high-reflective interference image, over-segmentation results will occur during the image segmentation process, and the punching area cannot be effectively and finely segmented, and the finally segmented punching area is incomplete, thus affecting the detection accuracy of the punching defects. The present invention analyzes the high-reflective characteristics of the punching on the inner tank plate of the washing machine, and performs adaptive marking on the area with punching characteristics before using the watershed algorithm to segment the image, obtaining a better segmentation effect.

[0054] Based on the above analysis, for the gray-scale image of the inner liner plate punching, an edge detection algorithm is used to obtain the edge lines in the gray-scale image of the inner liner plate punching, and the pixel points on the closed edge lines are used as the closed edge pixels.

[0055] For the closed edge line, calculate the circularity of the area enclosed by the closed edge line. The closer the value of the circularity is to the numerical value 1, the closer the area enclosed by the closed edge line is to a circle. Set a preset circularity threshold. In an embodiment of the present invention, the value of the preset circularity threshold is 0.9, and the implementer can select other values according to the actual situation.

[0056] Specifically, the punching holes on the washing machine inner liner plate are standard circles under normal circumstances. However, due to the relatively smooth surface of the washing machine inner liner plate, reflective images will be generated, which will affect the division of the punching hole area. Therefore, for high-reflection interference, specifically:

[0057] In the inner liner plate punching image, the area with a higher circularity may be the punching hole area on the washing machine inner liner plate or the reflective image on the inner liner plate. Based on the above analysis, the area enclosed by the closed edge line with a circularity greater than the preset circularity threshold is used as an approximate circle area, and the area with a circularity greater than the preset circularity threshold is analyzed for features.

[0058] In this embodiment, by collecting the gray-scale image of the inner liner plate punching of the washing machine, the state of the inner liner plate punching can be intuitively displayed, so as to improve the accuracy of detecting the defects of the inner liner plate punching.

[0059] Step S2: Divide the approximate circle area into a suspected light spot area and a light ring area.

[0060] Specifically, when taking an image of the punching hole on the washing machine inner liner plate, there is a certain arc between the punching hole and the inner liner plate, and there is no direct cut. Therefore, an aperture or a light spot will appear near the punching hole.

[0061] Therefore, it is necessary to analyze the aperture and light spot distribution characteristics of the pixel points of the punching hole on the washing machine inner liner plate. The specific method is as follows:

[0062] As Figure 2 shown, for the edge pixels of each approximate circle area in the gray-scale image of the inner liner plate punching, obtain the smallest circumscribed circle that includes the approximate circle area, and obtain a preset number of consecutive peripheral circles adjacent to the smallest circumscribed circle. In an embodiment of the present invention, the value of the preset number is 3, and the implementer can select other values according to the actual situation.

[0063] An aperture feature will appear near the punching hole. Since the color inside the punching hole is darker, and since the edge of the punching hole is not a sharp edge but a gradually transitional arc edge with a slight bulge, therefore, the brightness difference of the aperture in the bulging part of the punching hole edge area is relatively large compared to the brightness inside the punching hole. That is, there will be a situation where the gray values of the pixel points in the aperture part of the bulging part of the punching hole edge area are extremely large, and the gray values of the pixel points inside the punching hole are extremely small.

[0064] For each approximate circular region in the gray image of the inner liner plate, use the HOG algorithm to obtain the histogram of gradient directions of all pixel points on the minimum circumscribed circle of the approximate circular region, and the histogram of gradient directions of all pixel points on the peripheral circle adjacent to the minimum circumscribed circle region.

[0065] As Figure 3 shown, the histogram of gradient directions includes the gradient amplitude and gradient direction of the gray value of the pixel points; perform a linear fitting process on the values of all bins (bins, intervals in the histogram of gradient directions) in the histogram of gradient directions, and obtain the fitting line of the histogram of gradient directions of the minimum circumscribed circle as the minimum fitting line of each approximate circular region, and the histogram of gradient directions of the peripheral circle adjacent to the minimum circumscribed circle region as the minimum adjacent fitting line of each approximate circular region.

[0066] In an embodiment of the present invention, the selected linear fitting algorithm is the Random Sample Consensus (RANSAC) algorithm, and the implementer can select other linear fitting algorithms according to the actual situation; among them, the HOG algorithm (Histogram of Oriented Gradients) is a method for feature description in the fields of computer vision and image processing, mainly used for object detection and image recognition tasks.

[0067] Furthermore, in order to reflect the degree to which the features in the area near the punching hole conform to the aperture reflection feature in the gray image, based on the average situation, dispersion degree of the gray values of the pixel points on the minimum circumscribed circle and peripheral circle of the approximate circular region, and the absolute value of the slope of the minimum fitting line and the minimum adjacent fitting line, calculate the aperture feature coefficient of each approximate circular region. The calculation formula is:

[0068] ;

[0069] where, hfc is the aperture feature coefficient of each approximate circular region; amp min is the mean value of the gradient amplitudes of all pixel points on the minimum circumscribed circle of each approximate circular region, amp cir is the mean value of the gradient amplitudes of all pixel points on the peripheral circle adjacent to the minimum circumscribed circle of each approximate circular region, var gra is the variance of the gray values of all pixel points on all peripheral circles of each approximate circular region, k minis the absolute value of the slope of the minimum fitting line for each approximate circular region, k cir is the absolute value of the slope of the minimum adjacent fitting line for each approximate circular region; is a preset parameter adjustment coefficient. To prevent the denominator from being a numerical value of 0, in one embodiment of the present invention the value is 1, and the implementer can select other values according to the actual situation; sigmoid() is a normalization function.

[0070] It should be noted that for each approximate circular region in the gray-scale image of the inner liner plate punching, when the surrounding region of the approximate circular region conforms to the halo characteristics during reflection, since the region inside the punching is darker in color, that is, the gray-scale value of the pixel points in the region inside the punching is smaller. At this time, if the approximate circular region is the punching region, the minimum circumscribed circle is the edge line of the punching region, the gray-scale value of the pixel points is smaller, and the outer circle of the minimum circumscribed circle is the reflective region of the inner liner plate punching edge. At this time, the pixel points on the outer circle are lighter in color, that is, the gray-scale value is larger, that is, the difference between the gray-scale value of the pixel points on the minimum circumscribed circle and the gray-scale value of the pixel points on its adjacent outer circle is larger. At this time, the obtained amp min 、amp cir has a larger value; at the same time, since the dividing line of the halo at the punching edge is relatively obvious, that is, there will be a large mutation between the gray-scale value of the pixel points on the minimum circumscribed circle and the gray-scale value of the pixel points on its adjacent outer circle. At this time, the gradient amplitude and gradient direction distribution of the pixel points on the minimum circumscribed circle and its adjacent outer circle are relatively uniform (the gradient amplitude is mainly obtained from the difference in the gray-scale value of the pixel points from the minimum circumscribed circle to its adjacent outer circle, and the gradient direction is uniformly distributed from 0 degrees to 360 degrees for the circle). The obtained gradient direction histogram is relatively smooth. At this time, the minimum fitting line and the minimum adjacent fitting line obtained are closer to the horizontal line, that is, the absolute values of the slopes k min 、k cir are smaller, and the value of the obtained halo feature coefficient hfc is larger at this time; on the contrary, when the surrounding region of the approximate circular region does not conform to the halo characteristics during reflection, the value of the obtained halo feature coefficient hfc is smaller at this time.

[0071] It should be noted that in one embodiment of the present invention, the preset value of the halo feature threshold is 0.8, and the implementer can select other values according to the actual situation, which is not specifically limited in the present invention. The approximate circular region with a halo feature coefficient less than the preset halo feature threshold is used as the suspected light spot region, and the region greater than or equal to the halo feature coefficient is used as the halo region.

[0072] Furthermore, through the analysis of the approximate circular region, the approximate circular region is divided. The approximate circular region with a halo feature coefficient less than the halo feature threshold is used as the suspected light spot region, and the approximate circular region greater than or equal to the halo feature threshold is used as the halo region;

[0073] In this embodiment, the accuracy of halo feature recognition is improved by calculating the halo feature coefficients of each approximate circular region.

[0074] Step S3: Calculate the spot feature coefficients of each suspected spot region, and use the suspected spot regions with spot feature coefficients greater than or equal to the spot feature threshold as spot regions. Take any one of the halo region and the spot region as the target region; calculate the in-hole feature coefficient of the target region.

[0075] Specifically, for the inner liner plate of the washing machine, when the spot feature appears on the inner liner plate of the washing machine under normal conditions, since the edge of the punching has a certain arc, there will be a situation where the spot appears in the punching area in the grayscale image of the inner liner plate punching. For the suspected spot regions in the grayscale image of the inner liner plate punching, the gray-level co-occurrence matrix is used to process the suspected spot regions to obtain the contrast of the suspected spot regions as the light and dark contrast of each suspected spot region; at the same time, in order to reflect the dark region in the minimum circumscribed circle of the suspected spot region and the bright points in the dark region, the SIFT algorithm is used to process the pixel points in each suspected spot region to obtain the feature vector of the minimum circumscribed circle of each suspected spot region. Further, the center point of the minimum circumscribed circle of each suspected spot region is obtained, and a circular region is constructed with the center point of the minimum circumscribed circle of each suspected spot region as the center as the neighborhood of the minimum circumscribed circle of each suspected spot region. Among them, in one embodiment of the present invention, the number of minimum circumscribed circles of the suspected spot regions included in the circular region is N, and the value of N in one embodiment of the present invention is 10, and the implementer can select other values according to the actual situation. Among them, the gray-level co-occurrence matrix is a method for describing texture by studying the spatial correlation characteristics of gray levels in an image. The gray-level co-occurrence matrix is the probability that the gray value is j at a point that departs from a pixel point with gray level i and is at a fixed position (the distance is d and the orientation is θ). All the estimated values in the gray-level co-occurrence matrix can be represented in the form of a matrix. The SIFT algorithm (Scale-Invariant Feature Transform) is an algorithm for image feature extraction and matching.

[0076] Further, when light spot features appear on the inner tank plate of the washing machine under normal conditions, the positions of the light spots in each punching hole and other punching holes around are relatively consistent. Based on the above analysis, the eigenvectors of the minimum circumscribed circles of each suspected light spot area are used as each feature point, and clustering processing is performed on all feature points in the neighborhood of the minimum circumscribed circles of each suspected light spot area to obtain each clustering cluster. The clustering algorithm selected in an embodiment of the present invention is the K-means clustering algorithm, and the implementer can select other clustering algorithms according to the actual situation; in order to reflect the consistency of the positions where the minimum circumscribed circle light spots appear in each suspected light spot area, a light spot consistency index of each clustering cluster is constructed, and the calculation formula is:

[0077] ;

[0078] where, ind fac is the light spot consistency coefficient of each clustering cluster; average() is the mean value function, is the set composed of the silhouette coefficients of all feature points in each clustering cluster. Among them, the silhouette coefficient is an index used to evaluate the quality of clustering. It evaluates the clustering effect by quantifying the similarity between data points and the clusters they belong to and other clusters.

[0079] It should be noted that when the minimum circumscribed circle of the suspected light spot area conforms more to the light spot features of the punching area under normal conditions, the positions where the light spots appear in the minimum circumscribed circle of the suspected light spot area and the minimum circumscribed circles of other suspected light spot areas in its neighborhood are consistent. At this time, the similarity between the eigenvectors of the minimum circumscribed circle of the suspected light spot area and the minimum circumscribed circles of other suspected light spot areas in its neighborhood is relatively high, the clustering effect of the clustering cluster is better, that is, the values of the silhouette coefficients of all feature points in the clustering cluster are larger, and the light spot consistency coefficient ind fac obtained at this time is larger; on the contrary, the obtained light spot consistency coefficient ind fac is smaller.

[0080] Further, based on the light spot consistency coefficient and the light and dark contrast ratio, a light spot feature coefficient of each approximate circle area is obtained, and the calculation formula is:

[0081] ;

[0082] where, ffc is the light spot feature coefficient of each suspected light spot area; ind fac is the light spot consistency coefficient of the clustering cluster where the minimum circumscribed circle of each suspected light spot area is located, ldc is the light and dark contrast ratio of the minimum circumscribed circle of each suspected light spot area; sigmoid() is the normalization function.

[0083] It should be noted that for the minimum circumscribed circle of each suspected light spot area, when the suspected light spot area conforms to the light spot characteristics of high reflectivity in punching, the value of the light spot consistency coefficient of the clustering cluster where the minimum circumscribed circle of the suspected light spot area is located is relatively large, and the brightness difference between the brighter area and the darker area in the minimum circumscribed circle of the suspected light spot area is relatively large, that is, the value of the brightness contrast is relatively large. At this time, the value of the obtained light spot feature coefficient is relatively large; on the contrary, the value of the obtained light spot feature coefficient is relatively small. Obtain the light spot feature coefficients of all suspected light spot areas, and set a preset light spot feature threshold. In one embodiment of the present invention, the value of the preset light spot feature threshold is 0.8, and the implementer can select other values according to the actual situation; for each suspected light spot area, use the suspected light spot area whose light spot feature coefficient is greater than or equal to the preset light spot feature threshold as the light spot area.

[0084] Further, taking any area in the halo area and the light spot area as the target area, for the target area in the inner liner punching grayscale image, calculate the variance of the grayscale values of all pixel points in the target area. In order to reflect the possibility that the distribution of the grayscale values of the pixel points in the target area conforms to the punching characteristics, based on the degree of dispersion and average situation of the grayscale values of the pixel points in the target area around the average value, obtain the in-hole feature coefficient of the target area. The calculation formula is:

[0085] ;

[0086] where pfc is the in-hole feature coefficient of the target area; var cir is the variance of the grayscale values of all pixel points in the target area, ave is the average value of the grayscale values of all pixel points in the target area, ave def is the preset grayscale average value. In one embodiment of the present invention, the value of ave def is 5, and the implementer can select other values according to the actual situation; is the preset tuning factor. In order to prevent the denominator from being 0, the value in one embodiment of the present invention is 1, and the implementer can select other values according to the actual situation.

[0087] It should be noted that in the inner liner punching grayscale image, under normal circumstances, the pixel points of the punching part are darker in color, the corresponding grayscale values of the pixel points of the punching part are smaller, and the grayscale values of the pixel points of the punching part are relatively uniform; therefore, based on the above analysis, if the distribution characteristics of the grayscale values of the pixel points in the target area more meet the distribution characteristics of the grayscale values of the pixel points of the punching part, that is, the grayscale values of the pixel points in the target area are relatively uniform, at this time, the obtained var cirThe value is small, and the color of the pixel points in the target area is darker, that is, the gray value of the pixel points in the target area is small. The present invention selects a smaller preset gray mean value, and reflects the color depth of the gray value in the target area through the difference between the average situation of the gray values of the pixel points in the target area and the preset gray mean value. When the color of the pixel points in the target area is darker, it is more in line with the distribution characteristics of the gray values of the pixel points in the punched part, and the obtained The value is small, and at this time, the value of the hole feature coefficient pfc obtained is large.

[0088] In this embodiment, by obtaining the hole feature coefficient of the target area, the detection accuracy and efficiency are effectively improved, and the quality control of the inner liner plate punching is optimized, thereby improving the product reliability and safety.

[0089] Step S4: Take the target area with the hole feature coefficient greater than the hole feature threshold as the punching feature area. Centered on each punching feature area, segment the inner liner plate punching gray image, and perform defect detection on the segmented gray image to obtain punching defects.

[0090] Specifically, for each light spot area and each light ring area, take the area with the hole feature coefficient greater than the preset hole feature threshold as the punching feature area.

[0091] For the inner liner plate punching gray image, use the watershed algorithm based on mark to segment it. Take the area centered on each punching feature area with a preset radius as the marking area. Among them, the value of the preset radius should be less than the radius of the punching. In an embodiment of the present invention, the value of the preset radius is 2, and the implementer can select other values according to the actual situation; then segment the inner liner plate punching gray image to obtain a more refined segmented image, and use the template matching algorithm based on the image pyramid to perform defect detection on the segmented inner liner plate punching gray image to obtain the punching defects of the washing machine inner liner plate.

[0092] In this embodiment, the punching feature area is obtained through the hole feature coefficient, the light ring feature coefficient and the light spot feature coefficient, and the inner liner plate punching gray image is marked based on the punching feature area to improve the detection accuracy of the punching defects of the washing machine inner liner plate.

Claims

1. A method for detecting punching defects of the inner liner plate of a washing machine, characterized in that, Including: Obtain the closed edge line of the punched gray-scale image of the inner tank plate to obtain an approximate circular area in the area enclosed by the closed edge line; the approximate circular area is the area enclosed by the closed edge line with a circularity greater than a preset circularity threshold; Divide the approximate circular area into a suspected light spot area and a light ring area; Calculate the light spot feature coefficients of each suspected light spot area, and use the suspected light spot areas with light spot feature coefficients greater than or equal to the light spot feature threshold as the light spot areas; the light spot feature coefficients are positively correlated with the contrast and the light spot consistency coefficient, and the light spot consistency coefficient characterizes the consistency of the positions where the light spots appear in each suspected light spot area; Use any area of the light ring area and the light spot area as the target area; calculate the in-hole feature coefficient of the target area, and the in-hole feature coefficient is inversely correlated with the variance and mean of the gray values of all pixel points in the target area; Use the target areas with in-hole feature coefficients greater than the in-hole feature threshold as the punching feature areas; Centering on each punching feature area, segment the punched gray-scale image of the inner tank plate, and perform defect detection on the segmented gray-scale image to obtain punching defects.

2. The punching defect detection method for the inner liner plate of a washing machine according to claim 1, wherein, The process of the approximate circle region division includes: regarding the approximate circle region with the halo feature coefficient less than the halo feature threshold as the suspected light spot region, and the approximate circle region with the halo feature coefficient greater than or equal to the halo feature threshold as the halo region; the halo feature coefficient hfc is: ; where amp min is the mean value of the gradient amplitudes of all pixel points on the minimum circumscribed circle of each approximate circle region, and amp cir is the mean value of the gradient amplitudes of all pixel points on the peripheral circle adjacent to the minimum circumscribed circle of each approximate circle region, var gra is the variance of the gray values of all pixel points on all peripheral circles of each approximate circle region, k min is the absolute value of the slope of the minimum fitting line of each approximate circle region, and k cir is the absolute value of the slope of the minimum adjacent fitting line of each approximate circle region; is a preset tuning parameter, and Sigmoid() is a normalization function.

3. The method for detecting punching defects of the inner liner plate of a washing machine according to claim 2, characterized in that, The process of obtaining the minimum fitting line is specifically as follows: For the approximate circular area in the punched gray-scale image of the inner tank plate, use the HOG algorithm to obtain the gradient direction histogram of all pixel points on the minimum circumscribed circle of the approximate circular area; Perform linear fitting processing on the values in the interval of the gradient direction histogram to obtain the minimum fitting line.

4. A method for detecting punching defects of an inner liner plate of a washing machine according to claim 1, characterized in that, The in-hole feature coefficient is specifically calculated as: ； Among them, pfc is the in-hole feature coefficient of the target area; var cir is the variance of the gray values of all pixel points in the target area, ave is the mean of the gray values of all pixel points in the target area, ave def is the preset gray mean value; is the preset parameter adjustment factor.

5. A method for detecting punching defects of an inner liner plate of a washing machine according to claim 1, characterized in that, The light spot feature coefficient is: ； Among them, ffc is the spot feature coefficient of each suspected spot area; ind fac is the spot consistency coefficient of the clustering cluster where the minimum circumscribed circle of each suspected spot area is located, ldc is the light and dark contrast of the minimum circumscribed circle of each suspected spot area; Sigmoid() is the normalization function.

6. The punching defect detection method for the inner liner plate of a washing machine according to claim 5, characterized in that, The light spot consistency coefficient is specifically: ； Among them, ind fac is the spot consistency coefficient of the clustering cluster; average() is the average value function, is a set composed of the silhouette coefficients of all feature points in the clustering cluster. The clustering cluster is obtained by clustering all feature points in the neighborhood of the minimum circumscribed circle of the suspected spot area, taking the feature vector of the minimum circumscribed circle of the suspected spot area as the feature points. The feature vector is obtained by processing the pixel points in the minimum circumscribed circle of the suspected spot area using the SIFT algorithm.

7. A method for detecting punching defects of an inner liner plate of a washing machine according to claim 1, characterized in that, The process of obtaining the defects of the punched inner tank plate specifically includes: For the punched gray-scale image of the inner tank plate, use the watershed algorithm of mark to segment the punched gray-scale image of the inner tank plate, and use the area with a preset radius centered on each punching feature area as the marking area, where the value of the preset radius should be less than the radius of the punching; Perform image segmentation on the punched gray-scale image of the inner tank plate within the marking area to obtain a segmented image, and use the template matching algorithm based on the image pyramid to perform defect detection on the segmented punched gray-scale image of the inner tank plate to obtain the punching defects of the inner tank plate.

8. A method for detecting punching defects of an inner tank plate of a washing machine according to claim 1, characterized in that, The method for marking the punched gray-scale image of the inner tank plate includes: using a marking algorithm to mark the positions of the punching feature areas on the image.

9. A method for detecting punching defects of an inner liner plate of a washing machine according to claim 1, characterized in that The punched gray-scale image of the inner tank plate is obtained by using a high-pixel industrial camera to collect the punched image of the inner tank plate, and after using image denoising to perform denoising processing on the punched image of the inner tank plate, performing gray-scale processing on the denoised punched image of the inner tank plate, where the image denoising is performed using the mean filtering denoising algorithm.

Images