An intelligent recognition system for stain area recognition

Abstract

The present invention relates to the field of image recognition technology and is an intelligent recognition system for stain area recognition, including: an image acquisition module, an image processing module, and a cleaning parameter output module; the image acquisition module is used to capture multi-view global images of the stain area and perform noise reduction and grayscale processing on the multi-view global images; the image processing module performs image block processing on the processed multi-view global images to obtain a first image block and a second image block set, confirms the stain type through the first image block set, and enhances the local edge features of the second image block to obtain the stain boundary; the cleaning parameter output module is used to output cleaning parameters; the present invention solves the problem in the prior art that due to the complex and uneven texture features of the stain area, the stain recognition process requires a large amount of computing power resources, resulting in low recognition efficiency and low cleaning efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition and is an intelligent recognition system for stain area recognition. Background Art

[0002] In the existing stain cleaning system, the following technical problems exist: First, the cleaning efficiency is low. Due to the lack of effective stain area recognition and cleaning path planning, the existing system often needs to repeat operations multiple times during the cleaning process, resulting in a long cleaning cycle and large resource consumption. Second, the stain recognition is not accurate enough. When the existing image processing technology recognizes the stain types and boundaries, it is often interfered by factors such as environmental illumination and image noise, resulting in frequent misjudgment or missed judgment. In addition, the existing system has poor adaptability to floors of different materials and cannot dynamically adjust the cleaning parameters according to the characteristics of the floor material, affecting the cleaning effect; at the same time, the technology for extracting and enhancing the edge features of stains is relatively weak, the features of the stain boundary area are not obvious, and it is difficult to accurately divide the stain area, thus affecting the determination of the cleaning range and the cleaning effect. Summary of the Invention

[0003] The technical problem to be solved by the present invention is aimed at the problem in the prior art that due to the complex and uneven texture features of the stain area, the stain recognition process requires a large amount of computing power resources, resulting in low recognition efficiency. An intelligent recognition system for stain area recognition is proposed.

[0004] In order to achieve the above object, an intelligent recognition system for stain area recognition of the present invention includes:

[0005] An image acquisition module, an image processing module, and a cleaning parameter output module;

[0006] The image acquisition module is used to capture multi-view global images of the stain area and perform noise reduction and grayscale processing on the multi-view global images through parallel technology;

[0007] The image processing module is used to perform image block processing on the processed multi-view global images to obtain a first image block set and a second image block set, confirm the stain types through the first image block set; establish a historical object surface texture library, and confirm the object surface where each second image block in the second image block set is located, and enhance the local edge features of the second image blocks to obtain the stain boundary;

[0008] The cleaning parameter output module is used to import the stain types and the enhanced stain boundary into the cleaning parameter calculation strategy and output the cleaning parameters.

[0009] Specifically, the image acquisition module is configured with the following strategy:

[0010] S11: Arrange 5 high-resolution RGB cameras at the four corners and the top of the stain area, and simultaneously turn on all the high-resolution RGB cameras to take pictures of the stain area to obtain multi-view global images. Among them, the multi-view global images include: 4 top corner images and 1 top image;

[0011] S12: Remove the noise in the multi-view global images through the median filtering algorithm, and simultaneously convert the multi-view global images into grayscale images.

[0012] Specifically, the image processing module includes: an image block processing unit;

[0013] The image block processing unit configures the following strategy:

[0014] S21: Use the edge detection algorithm to identify the initial edge points of the stain area in the top image;

[0015] S22: Construct a Gaussian mixture model, represent the initial edge points as a mixture of multiple Gaussian distributions, and dynamically adjust the weights of each Gaussian distribution according to the edge curvature to calculate the edge point density of each pixel point;

[0016] Preferably, by dynamically adjusting the density, the density of edge points is higher in the area with larger curvature, thereby improving the image block accuracy;

[0017] Preferably, the calculation strategy of the edge point density is:

[0018] ;

[0019] Wherein, represents the edge point density of the pixel point with coordinates ;

[0020] K is the number of Gaussian distributions in the Gaussian mixture model, and k is the index of the Gaussian distribution;

[0021] is the weight of the k-th Gaussian distribution, indicating the contribution ratio of the k-th Gaussian distribution in all edge point distributions;

[0022] is the probability density function of the k-th Gaussian distribution, is the mean vector of the k-th Gaussian distribution; is the covariance matrix of the k-th Gaussian distribution;

[0023] represents the edge curvature weight at the pixel point with coordinates ;

[0024] S23: Extract the image gradient information, guide the direction of region growth based on the edge point density, image gradient information, and edge curvature, determine whether to add the next pixel point to the current region, traverse all pixel points, and obtain the first image block;

[0025] It should be noted that by guiding the direction of region growth based on the edge point density, image gradient information, and edge curvature, the growth of each region during the image segmentation process is more inclined to proceed along the direction of the edge or texture.

[0026] Preferably, determining whether to add the next pixel to the current region specifically includes:

[0027] ;

[0028] Among them, represents the judgment value for whether to add the pixel point with coordinates to the current region. When , add the pixel point with coordinates to the current region, otherwise do not add;

[0029] represents the gradient information of the pixel change of the pixel point with coordinates ;

[0030] is the edge curvature of the pixel point with coordinates ;

[0031] is the region growth judgment function;

[0032] S24: Extract the edge curvature of each edge point on the edge line of the first image block, preset the edge curvature threshold, and filter out the edge points greater than the edge curvature threshold as the segmentation points;

[0033] S25: Construct a segmentation line fitting model, input the filtered segmentation points into the segmentation line fitting model, fit and generate a segmentation line, and divide the edge region into multiple second image blocks according to the segmentation line to form a second image block set.

[0034] Specifically, the image processing module further includes: a stain type confirmation unit;

[0035] The stain type confirmation unit configures the following strategy:

[0036] S31: Use the SIFT (Scale-Invariant Feature Transform) algorithm to detect key pixel points in the multi-view global image, calculate the descriptors of all key pixel points, use the nearest neighbor matching to find the matching point pairs, and apply the bidirectional matching algorithm to remove the incorrect matches;

[0037] S32: Obtain the internal and external parameters of the camera through the calibration board. Use the matching point pairs in the multi-view global image and the internal and external parameters of the camera. According to the triangulation algorithm, calculate the spatial positions of each group of matching point pairs, combine the three-dimensional coordinates of all matching point pairs into point cloud data, and generate a three-dimensional surface model of the stain.

[0038] S33: By extracting geometric features, texture features, and color features, compare the geometric features, texture features, and color features with the historical stain features in the historical stain dataset, calculate the similarity, select the historical stain type with the highest similarity as the alternative stain type, and output the alternative stain type and the historical cleaning records of the alternative stain type.

[0039] Specifically, the image processing module further includes: an image block enhancement unit;

[0040] The image block enhancement unit includes a judgment subunit, and the judgment subunit is configured with the following strategy:

[0041] S41: Based on the coordinate position of the second image block in the global image, construct a feature vector using the feature information of the surrounding pixels.

[0042] S42: Construct a support vector machine classifier, and determine the ground property by judging the environment background where the second image block is located through the support vector machine classifier. The ground property includes: wood, metal, and plastic.

[0043] S43: If the confidence of the support vector machine classifier is lower than the preset classification threshold, the judgment of the ground property is uncertain. Extract the texture features of a larger area around the second image block, calculate the contrast, correlation, and energy of the texture using the gray-level co-occurrence matrix, and make a judgment again.

[0044] S43: Calculate the information entropy of the stain image, and at the same time calculate the gradient difference between the stain image and the surrounding background image. Preset the information entropy threshold and the gradient difference threshold. When the information entropy is less than the information entropy threshold and the gradient difference is less than the gradient difference threshold, the second image block needs to be enhanced to enhance the stain features and boundary clarity.

[0045] Specifically, the image block enhancement unit further includes an enhancement subunit, and the enhancement subunit is configured with the following strategy:

[0046] S44: After determining the ground property where the second image block is located, extract the corresponding texture template from the pre-established historical object surface texture library.

[0047] S45: Use the cross-correlation algorithm for texture matching.

[0048] Preferably, the cross-correlation algorithm is specifically:

[0049] ;

[0050] wherein, represents the cross - correlation coefficient, w is the window size; i and j respectively represent the offsets in the horizontal and vertical directions, ; I is the second image block, represents the pixel value of the pixel point of the second image block at the coordinate ; represents the pixel value of the pixel point of the texture template at the coordinate ;

[0051] are the means of the second image block and the texture template respectively;

[0052] S46: According to the matching result, adjust the contrast and brightness of the image through the adaptive histogram equalization algorithm. In the adaptive histogram equalization algorithm, the image is divided into multiple sub - blocks, and histogram equalization is performed on each sub - block image;

[0053] Preferably, the transformation function in the adaptive histogram equalization algorithm is: , wherein, is the transformed gray value, is the number of pixels with gray value q, n is the total number of pixels in the sub - block, and Q is the total number of gray values;

[0054] S47: During the matching process, if a texture part similar to the stain feature is found, adopt the local contrast enhancement algorithm to highlight the stain feature and suppress the interference of the background texture.

[0055] Specifically, the image block enhancement unit further includes a material compensation sub - unit, and the material compensation sub - unit is configured with a metal compensation strategy, a wood compensation strategy, and a plastic compensation strategy;

[0056] The metal compensation strategy includes: by analyzing the brightness distribution of the image , adopt the local illumination correction algorithm. First, calculate the average brightness of the local area , and then correct the brightness of each pixel through the correction formula to compensate for the brightness difference caused by reflection, making the stain feature more clearly distinguishable;

[0057] The correction formula is: , is the corrected pixel brightness, is the expected value of the pixel brightness;

[0058] It should be noted that the reflection on the metal surface will affect the recognition of stain features. Through the local illumination correction algorithm, the brightness difference caused by reflection can be compensated, making the stain features clearer and more distinguishable, eliminating the influence of the metal surface reflection on the stain features, and enhancing the visibility of the stains on the metal surface.

[0059] The described wood compensation strategy includes: filtering the image using a Gabor filter bank to highlight the texture features related to the stains, and at the same time using morphological operations to dilate the stain boundaries;

[0060] Specifically, the Gabor filter is a filter with direction selectivity and frequency selectivity. Its kernel function combines a Gaussian function and a cosine function. By adjusting the filter parameters, including wavelength, direction, phase, bandwidth, and aspect ratio, it filters the texture features in different directions and frequencies. The wood surface has directional and periodic texture features, and the Gabor filter can filter the image according to these features, highlighting the texture features related to the stains, effectively extracting the texture features of the wood surface, and enhancing the texture information related to the stains.

[0061] The described plastic compensation strategy includes: adopting an image diffusion model based on partial differential equations, adaptively adjusting according to the image gradient, and enhancing the boundaries of the stain regions through the diffusion process.

[0062] It should be noted that the image diffusion model, based on partial differential equations, describes the diffusion process of the image over time. During the diffusion process, the smooth regions of the image diffuse faster, and the edge regions diffuse slower. According to the adsorption characteristics of plastics to stains, using this model can enhance the boundaries of the stain regions while maintaining the smoothness of the image, avoiding image distortion caused by over-enhancement, making the stain features clearer and not affecting the image quality.

[0063] Specifically, the cleaning parameter output module includes: a cleaning curve fitting unit;

[0064] The cleaning curve fitting unit is configured with a cleaning parameter calculation strategy, specifically including:

[0065] S51: Select the midpoint of the curve connecting the first image block and the second image block as the target point, and select the point farthest from the target point from the points on the stain boundary in the enhanced second image block as the representative point. Traverse the set of second image blocks to obtain a set of representative points, where the total number of representative points in the set of representative points is B, and b is the representative point index;

[0066] S52: Set the circular domain equation, solve the center coordinates and radius, and calculate the sum of the squared circular domain errors of each representative point in the set of representative points with respect to the obtained circle;

[0067] Preferably, to find the center of the circle, it is necessary to minimize the total squared distance from the representative points to the center of the circle, that is, to minimize the following objective function: , where are the abscissa and ordinate of the representative point with index b in the set of representative points, are the coordinates of the center of the circle;

[0068] The equation of the circular domain is specifically:

[0069] ;

[0070] where is the sum of the x - coordinates of all representative points; is the sum of the y - coordinates of all representative points;

[0071] is the sum of the squares of the x - coordinates of the representative points; is the sum of the squares of the y - coordinates of the representative points;

[0072] The sum of the products of the x - coordinates and y - coordinates of all representative points; S is the sum of the squares of the x - coordinates and the squares of the y - coordinates of all representative points;

[0073] Solve the equation of the circular domain by Gaussian elimination or matrix inversion to obtain the coordinates of the center of the circle ;

[0074] The first equation is obtained based on the minimization of The second set of equations: To consider the symmetry and skewness in the distribution of points, is introduced;

[0075] The formula for calculating the radius is: , where r is the radius, are the abscissa and ordinate of the representative point with index b in the set of representative points;

[0076] The sum of the squared errors of the circular domain is calculated as:

[0077] ;

[0078] S53: Set up the equation of the elliptical domain, solve for the coordinates of the center of the ellipse, the major - axis radius, and the minor - axis radius, and calculate the sum of the squared errors of the elliptical domain of each representative point in the set of representative points with respect to the obtained ellipse;

[0079] Preferably, the equation of the elliptical domain is specifically:

[0080] ;

[0081] where is the center of the ellipse;

[0082] The calculation formulas for the major axis radius a and the minor axis radius z are as follows:

[0083] ;

[0084] The sum of squared errors of the elliptical domain The calculation formula is as follows:

[0085] ;

[0086] S54: Compare the sum of squared errors of the circular domain with the sum of squared errors of the elliptical domain. When the sum of squared errors of the circular domain is less than the sum of squared errors of the elliptical domain, select the obtained circular curve as the cleaning curve; otherwise, select the obtained elliptical curve as the cleaning curve.

[0087] Compared with the prior art, the technical effects of the present invention are as follows:

[0088] 1. By intelligently planning the cleaning curve, the present invention reduces the number of cleaning times and repetitive operations, significantly improves the cleaning efficiency, and shortens the cleaning cycle.

[0089] 2. Utilizing multi-view image acquisition, image block processing, and feature extraction technologies, the present invention realizes accurate identification of stain types and boundaries, reduces misjudgment and missed judgment phenomena, and improves the cleaning effect.

[0090] 3. According to the characteristics of different materials, the present invention dynamically adjusts cleaning parameters, optimizes cleaning strategies, ensures efficient and accurate cleaning on surfaces of different materials, and improves the adaptability and robustness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for description in the embodiments. Obviously, the following described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. Among them:

[0092] Figure 1 is a schematic structural diagram of an intelligent recognition system for stain area recognition of the present invention;

[0093] Figure 2 is a schematic structural diagram of an image processing module of the present invention;

[0094] Figure 3 is a schematic diagram of the principle of region growing of the present invention;

[0095] Figure 4Schematic diagram of a scenario for processing multi - perspective global images of stains by an image processing module according to the present invention. Detailed implementation manners

[0096] To make the above - mentioned objects, features and advantages of the present invention more obvious and understandable, the following will describe the detailed implementation manners of the present invention in conjunction with the accompanying drawings of the specification.

[0097] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0098] Secondly, the so - called "one embodiment" or "embodiment" herein refers to a specific feature, structure or characteristic that can be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or selectively exclusive embodiments from other embodiments.

[0099] Embodiment 1:

[0100] As Figures 1-4 shown, an intelligent recognition system for stain area recognition according to an embodiment of the present invention, as Figure 1 shown, includes the following modules:

[0101] An image acquisition module, an image processing module, and a cleaning parameter output module;

[0102] The image acquisition module is used to capture multi - perspective global images of the stain area and perform noise reduction and grayscale processing on the multi - perspective global images through parallel technology;

[0103] The image acquisition module is configured with the following strategy:

[0104] S11: Arrange 5 high - resolution RGB cameras at the four corners and the top of the stain area, and simultaneously turn on all the high - resolution RGB cameras to photograph the stain area to obtain multi - perspective global images. Among them, the multi - perspective global images include: 4 top - corner images and 1 top image;

[0105] S12: Remove the noise in the multi - perspective global images through the median filtering algorithm, and at the same time convert the multi - perspective global images into grayscale images.

[0106] The image processing module is used to perform image block processing on the processed multi-view global image to obtain a first set of image blocks and a second set of image blocks, confirm the stain types through the first set of image blocks; establish a historical object surface texture library, and confirm the object surface where each second image block in the second set of image blocks is located, enhance the local edge features of the second image blocks to obtain the stain boundaries.

[0107] The image processing module includes: an image block processing unit;

[0108] The image block processing unit configures the following strategy:

[0109] S21: Use an edge detection algorithm to identify the initial edge points of the stain area in the top image;

[0110] S22: Construct a Gaussian mixture model, represent the initial edge points as a mixture of multiple Gaussian distributions, and dynamically adjust the weights of each Gaussian distribution according to the edge curvature to calculate the edge point density of each pixel point;

[0111] Preferably, by dynamically adjusting the density, the density of edge points is higher in areas with larger curvature, thereby improving the image block accuracy;

[0112] Preferably, the calculation strategy of the edge point density is:

[0113] ;

[0114] Where, represents the edge point density of the pixel point with coordinates ;

[0115] K is the number of Gaussian distributions in the Gaussian mixture model, and k is the index of the Gaussian distribution;

[0116] is the weight of the k-th Gaussian distribution, representing the contribution ratio of the k-th Gaussian distribution in the distribution of all edge points;

[0117] is the probability density function of the k-th Gaussian distribution, is the mean vector of the k-th Gaussian distribution; is the covariance matrix of the k-th Gaussian distribution;

[0118] represents the edge curvature weight at the pixel point with coordinates ;

[0119] Exemplarily, in this embodiment, , where, is the edge curvature of the pixel point with coordinates .

[0120] S23: Extract the image gradient information, guide the direction of region growing according to the edge point density, the image gradient information, and the edge curvature, determine whether to add the next pixel point to the current region, traverse all pixel points, and obtain the first image block;

[0121] It should be noted that by guiding the direction of region growing according to the edge point density, the image gradient information, and the edge curvature, the growth of each region during the image block division process is more inclined to proceed along the edge or texture direction.

[0122] Preferably, determining whether to add the next pixel to the current region specifically includes:

[0123] ;

[0124] Among them, represents the judgment value for determining whether to add the pixel point with coordinates to the current region. When , add the pixel point with coordinates to the current region; otherwise, do not add it.

[0125] represents the gradient information of the pixel change of the pixel point with coordinates ;

[0126] is the edge curvature of the pixel point with coordinates ;

[0127] is the region growing judgment function;

[0128] Exemplarily, in this embodiment, A, B, and C are respectively three judgment quantities in the region growing judgment function;

[0129] ;

[0130] Among them, represents the dynamic threshold, which is a linear combination of the gradient information and the edge curvature. are respectively the proportion coefficients of the gradient information and the edge curvature; by adjusting the value of , the influence degrees of the gradient and the curvature on the threshold can be controlled. In this way, the region growing process can be dynamically adjusted according to the local features of the image, which is more flexible and robust than the existing region growing algorithms and can handle the complex stain images corresponding to more types of stain types.

[0131] S24: Extract the edge curvature of each edge point on the edge line of the first image block, preset the edge curvature threshold, and screen the edge points greater than the edge curvature threshold as the segmentation points;

[0132] S25: Construct a dividing line fitting model, input the screened dividing points into the dividing line fitting model, fit and generate a dividing line, and divide the edge region into multiple second image patches according to the dividing line to form a second image patch set.

[0133] The image processing module further includes: a stain type confirmation unit;

[0134] The stain type confirmation unit configures the following strategy:

[0135] S31: Use the SIFT (Scale-Invariant Feature Transform) algorithm to detect key pixel points in the multi-view global image, calculate the descriptors of all key pixel points, use the nearest neighbor matching to find matching point pairs, and apply the bidirectional matching algorithm to remove incorrect matches;

[0136] S32: Obtain the internal and external parameters of the camera through the calibration board, use the matching point pairs in the multi-view global image and the internal and external parameters of the camera, calculate the spatial positions of each group of matching point pairs according to the triangulation algorithm, combine the three-dimensional coordinates of all matching point pairs into point cloud data, and generate a three-dimensional surface model of the stain;

[0137] S33: By extracting geometric features, texture features and color features, compare the geometric features, texture features and color features with the historical stain features in the historical stain dataset, calculate the similarity, select the historical stain type with the highest similarity as the alternative stain type, and output the alternative stain type and the historical cleaning record of the alternative stain type.

[0138] The image processing module further includes: an image patch enhancement unit;

[0139] The image patch enhancement unit includes a judgment sub-unit, and the judgment sub-unit configures the following strategy:

[0140] S41: Based on the coordinate position of the second image patch in the global image, construct a feature vector using the feature information of the surrounding pixels;

[0141] S42: Construct a support vector machine classifier, and determine the environmental background of the second image patch through the support vector machine classifier to determine the ground property, where the ground property includes: wood, metal and plastic;

[0142] S43: If the confidence level of the support vector machine classifier is lower than the preset classification threshold, the ground property judgment is uncertain, extract the texture features of a larger area around the second image patch, calculate the contrast, correlation and energy of the texture using the gray-level co-occurrence matrix, and make a judgment again;

[0143] S43: Calculate the information entropy of the stain image, and at the same time calculate the gradient difference between the stain image and the surrounding background image. Preset the information entropy threshold and the gradient difference threshold. When the information entropy is less than the information entropy threshold and the gradient difference is less than the gradient difference threshold, the second image block needs to be enhanced to enhance the stain features and boundary sharpness.

[0144] The image block enhancement unit further includes an enhancement subunit, and the enhancement subunit is configured with the following strategy:

[0145] S44: After determining the ground property where the second image block is located, extract the corresponding texture template from the pre-established historical object surface texture library;

[0146] S45: Use the cross-correlation algorithm for texture matching;

[0147] Preferably, the cross-correlation algorithm is specifically:

[0148] ;

[0149] where, represents the cross-correlation coefficient, w is the window size; i and j respectively represent the offsets in the horizontal and vertical directions, ; I is the second image block, represents the pixel value of the pixel point of the second image block at the coordinate ; represents the pixel value of the pixel point of the texture template at the coordinate ;

[0150] are the means of the second image block and the texture template respectively;

[0151] S46: According to the matching result, adjust the contrast and brightness of the image through the adaptive histogram equalization algorithm. In the adaptive histogram equalization algorithm, the image is divided into multiple sub-blocks, and histogram equalization is performed on each sub-block image;

[0152] Preferably, the transformation function in the adaptive histogram equalization algorithm is: , where, is the transformed gray value, is the number of pixels with gray value q, n is the total number of pixels in the sub-block, and Q is the total number of gray values;

[0153] S47: During the matching process, if a texture part similar to the stain feature is found, use the local contrast enhancement algorithm to highlight the stain feature and suppress the interference of the background texture.

[0154] Exemplarily, in this embodiment, the local contrast enhancement algorithm is specifically:

[0155] , is the enhancement coefficient, is the average value of neighboring pixels; are the pixels of the second image block before and after enhancement respectively;

[0156] The image block enhancement unit further includes a material compensation subunit, and the material compensation subunit is configured with a metal compensation strategy, a wood compensation strategy, and a plastic compensation strategy;

[0157] The metal compensation strategy includes: by analyzing the brightness distribution of the image , using a local illumination correction algorithm, first calculating the average brightness of the local area , and then correcting the brightness of each pixel through a correction formula to compensate for the brightness difference caused by reflection, making the stain features more clearly distinguishable;

[0158] The correction formula is: , is the corrected pixel brightness, is the expected value of the pixel brightness.

[0159] It should be noted that the reflection on the metal surface will affect the recognition of stain features. Through the local illumination correction algorithm, the brightness difference caused by reflection can be compensated, making the stain features more clearly distinguishable, eliminating the influence of the reflection on the metal surface on the stain features, and enhancing the visibility of the stains on the metal surface.

[0160] The wood compensation strategy includes: filtering the image using a Gabor filter bank to highlight the texture features related to the stains, and at the same time using morphological operations to dilate the stain boundaries;

[0161] Specifically, the Gabor filter is a filter with direction selectivity and frequency selectivity. Its kernel function combines a Gaussian function and a cosine function. By adjusting the filter parameters, including wavelength, direction, phase, bandwidth, and aspect ratio, it filters the texture features in different directions and frequencies. The wood surface has directional and periodic texture features, and the Gabor filter can filter the image according to these features to highlight the texture features related to the stains, effectively extract the texture features of the wood surface, and enhance the texture information related to the stains.

[0162] The plastic compensation strategy includes: adopting an image diffusion model based on partial differential equations, adaptively adjusting according to the image gradient, and enhancing the boundaries of the stain regions through the diffusion process.

[0163] Exemplarily, in this embodiment, the image diffusion model based on partial differential equations is specifically: , where \(N\) represents an image function, which is a function of spatial coordinates and time \(t\), and describes the pixel values of the image at different positions at different times; represents the partial derivative of the image function \(N\) with respect to time \(t\), is the gradient operator, is the divergence operator, is used to calculate the divergence of the vector field to control the direction and intensity of image diffusion; is the diffusion coefficient;

[0164] It should be noted that the image diffusion model, based on partial differential equations, describes the diffusion process of the image over time. During the diffusion process, the smooth regions of the image diffuse faster, and the edge regions diffuse slower. According to the adsorption characteristics of plastics to stains, using this model can enhance the boundary of the stain region while maintaining the smoothness of the image, avoiding image distortion caused by over-enhancement, making the stain features clearer and the image quality unaffected.

[0165] The cleaning parameter output module is used to import the stain type and the enhanced stain boundary into the cleaning parameter calculation strategy and output the cleaning parameters.

[0166] The cleaning parameter output module includes: a cleaning curve fitting unit;

[0167] The cleaning curve fitting unit is configured with a cleaning parameter calculation strategy, specifically including:

[0168] S51: Select the midpoint of the curve connecting the first image block and the second image block as the target point, and screen the point farthest from the target point from the points on the stain boundary in the enhanced second image block as the representative point. Traverse the second image block set to obtain the representative point set, where the total number of representative points in the representative point set is \(B\), and \(b\) is the representative point index;

[0169] S52: Set the circular domain equation, solve the center coordinates and radius, and calculate the sum of the squared circular domain errors of each representative point in the representative point set with respect to the obtained circle;

[0170] Preferably, in order to find the center of the circle, it is necessary to minimize the total squared distance from the representative points to the center of the circle, that is, to minimize the following objective function: , where, are the abscissa and ordinate of the representative point with index \(b\) in the representative point set, are the center coordinates of the circle;

[0171] The circular domain equation is specifically:

[0172] ;

[0173] where, is the sum of the x - coordinates of all representative points; is the sum of the y - coordinates of all representative points;

[0174] is the sum of the squares of the x - coordinates of the representative points; is the sum of the squares of the y - coordinates of the representative points;

[0175] The sum of the products of the x - coordinates and y - coordinates of all representative points; S is the sum of the squares of the x - coordinates and the squares of the y - coordinates of all representative points;

[0176] Solve the circular domain equation by Gaussian elimination or matrix inversion to obtain the center coordinates of the circle ;

[0177] The first equation is obtained based on the minimization of The second set of equations: To consider the symmetry and skewness in the distribution of points, is introduced;

[0178] The calculation formula for the radius is: , where r is the radius, are the abscissa and ordinate of the representative point with index b in the representative point set;

[0179] The sum of squared errors of the circular domain The calculation formula is:

[0180] ;

[0181] S53: Set up the elliptical domain equation, solve for the center coordinates of the ellipse, the major - axis radius, and the minor - axis radius, and calculate the sum of squared errors of the elliptical domain of each representative point in the representative point set with respect to the obtained ellipse;

[0182] Preferably, the elliptical domain equation is specifically:

[0183] ;

[0184] where, is the center of the ellipse;

[0185] The calculation formulas for the major - axis radius a and the minor - axis radius z are:

[0186] ;

[0187] The sum of squared errors of the elliptical domain The calculation formula is:

[0188] ;

[0189] S54: Compare the sum of squared circle domain errors with the sum of squared ellipse domain errors. When the sum of squared circle domain errors is less than the sum of squared ellipse domain errors, select the obtained circular curve as the cleaning curve; otherwise, select the obtained elliptical curve as the cleaning curve.

[0190] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in the present invention can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0191] Those skilled in the art can clearly understand that for the sake of convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0192] In several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only one way, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms.

[0193] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0194] In addition, the functional units in each embodiment of the present invention can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0195] In the description of this specification, the descriptions referring to terms such as "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0196] In summary of the above embodiments, the technical effects of the present invention are as follows:

[0197] 1. By intelligently planning the cleaning curve, the present invention reduces the number of cleaning times and repetitive operations, significantly improves the cleaning efficiency, and shortens the cleaning cycle.

[0198] 2. By using multi-view image acquisition, image block processing and feature extraction technologies, the present invention realizes accurate identification of the types and boundaries of stains, reduces misjudgment and missed judgment phenomena, and improves the cleaning effect.

[0199] 3. According to the characteristics of different materials, the present invention dynamically adjusts the cleaning parameters, optimizes the cleaning strategy, ensures efficient and accurate cleaning on the surfaces of different materials, and improves the adaptability and robustness of the system.

[0200] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent recognition system for stain area recognition, characterized in that, The intelligent recognition system includes: an image acquisition module, an image processing module, and a cleaning parameter output module; The image acquisition module is used to capture multi-view global images of the stain area and perform noise reduction and grayscale processing on the multi-view global images through parallel technology; The image processing module is used to perform image block processing on the processed multi-view global images to obtain a first image block set and a second image block set, confirm the stain type through the first image block set; establish a historical object surface texture library, and confirm the object surface where each second image block in the second image block set is located, enhance the local edge features of the second image blocks to obtain the stain boundary; The cleaning parameter output module is used to import the stain type and the enhanced stain boundary into the cleaning parameter calculation strategy and output the cleaning parameters; The image acquisition module is configured with the following strategy: S11: Arrange 5 high-resolution RGB cameras at the four corners and the top of the stain area, and simultaneously turn on all the high-resolution RGB cameras to take pictures of the stain area to obtain multi-view global images. Among them, the multi-view global images include: 4 top corner images and 1 top image; S12: Remove the noise in the multi-view global images through the median filtering algorithm, and at the same time convert the multi-view global images into grayscale images; The image processing module includes: an image block processing unit; The image block processing unit is configured with the following strategy: S21: Use the edge detection algorithm to identify the initial edge points of the stain area in the top image; S22: Construct a Gaussian mixture model, represent the initial edge points as a mixture of multiple Gaussian distributions, and dynamically adjust the weights of each Gaussian distribution according to the edge curvature to calculate the edge point density of each pixel point; S23: Extract the image gradient information, determine the direction of region growth according to the edge point density, image gradient information, and edge curvature, and decide whether to add the next pixel point to the current region. Traverse all pixel points to obtain the first image block; S24: Extract the edge curvature of each edge point on the edge line of the first image block, preset the edge curvature threshold, and screen out the edge points greater than the edge curvature threshold as the segmentation points; S25: Construct a segmentation line fitting model, input the screened segmentation points into the segmentation line fitting model, fit and generate a segmentation line, and divide the edge region into multiple second image blocks according to the segmentation line to form a second image block set; The cleaning parameter output module includes: a cleaning curve fitting unit; The cleaning curve fitting unit is configured with a cleaning parameter calculation strategy, which specifically includes: S51: Select the midpoint of the curve connecting the first image block and the second image block as the target point, and screen out the point farthest from the target point from the points on the stain boundary in the enhanced second image block as the representative point. Traverse the second image block set to obtain a representative point set; S52: Set the circular domain equation, solve the center coordinates and radius, and calculate the sum of the squared circular domain errors of each representative point in the representative point set and the obtained circle; S53: Set the elliptical domain equation, solve for the center coordinates, major axis radius, and minor axis radius of the ellipse, and calculate the sum of the squared errors of the elliptical domain for each representative point in the set of representative points; S54: Compare the sum of the squared errors of the circular domain with the sum of the squared errors of the elliptical domain. When the sum of the squared errors of the circular domain is less than the sum of the squared errors of the elliptical domain, select the circular curve obtained by the solution as the cleaning curve; otherwise, select the elliptical curve obtained by the solution as the cleaning curve.

2. The intelligent recognition system for stain area recognition according to claim 1, wherein, The image processing module further includes: a stain type confirmation unit; The stain type confirmation unit is configured with the following strategy: S31: Use the SIFT algorithm to detect key pixel points in the multi-view global image, calculate the descriptors of all key pixel points, find the matching point pairs using the nearest neighbor matching, and apply the bidirectional matching algorithm to remove the incorrect matches; S32: Obtain the internal and external parameters of the camera through the calibration board. Use the matching point pairs in the multi-view global image and the internal and external parameters of the camera, calculate the spatial positions of each group of matching point pairs according to the triangulation algorithm, combine the three-dimensional coordinates of all matching point pairs into point cloud data, and generate the three-dimensional surface model of the stain; S33: By extracting geometric features, texture features, and color features, compare the geometric features, texture features, and color features with the historical stain features in the historical stain dataset, calculate the similarity, select the historical stain type with the highest similarity as the alternative stain type, and output the alternative stain type and the historical cleaning record of the alternative stain type.

3. The intelligent recognition system for stain area recognition according to claim 2, wherein The image processing module further includes: an image block enhancement unit; The image block enhancement unit includes a judgment sub-unit, and the judgment sub-unit is configured with the following strategy: S41: Based on the coordinate position of the second image block in the global image, construct a feature vector using the feature information of the surrounding pixels; S42: Construct a support vector machine classifier, and determine the ground attribute where the second image block is located through the support vector machine classifier. The ground attributes include: wood, metal, and plastic; S43: If the confidence level of the support vector machine classifier is lower than the preset classification threshold, the judgment of the ground attribute is uncertain. Extract the texture features of a larger area around the second image block, calculate the contrast, correlation, and energy of the texture using the gray-level co-occurrence matrix, and make a judgment again; S43: Calculate the information entropy of the stain image, and at the same time calculate the gradient difference between the stain image and the surrounding background image. Preset the information entropy threshold and the gradient difference threshold. When the information entropy is less than the information entropy threshold and the gradient difference is less than the gradient difference threshold, perform enhancement on the second image block to enhance the stain features and boundary clarity.

4. The intelligent recognition system for stain area recognition according to claim 3, characterized in that, The image block enhancement unit further includes an enhancement sub-unit, and the enhancement sub-unit is configured with the following strategy: S44: After determining the ground attribute where the second image block is located, extract the corresponding texture template from the pre-established historical object surface texture library; S45: Use the cross-correlation algorithm for texture matching; S46: According to the matching result, adjust the contrast and brightness of the image through the adaptive histogram equalization algorithm. In the adaptive histogram equalization algorithm, divide the image into multiple sub-blocks, and perform histogram equalization on each sub-block image; S47: During the matching process, if a texture part similar to the stain feature is found, a local contrast enhancement algorithm is adopted to highlight the stain feature.

5. The intelligent recognition system for stain area recognition according to claim 4, characterized in that The image block enhancement unit further includes a material compensation subunit, and the material compensation subunit is configured with a metal compensation strategy, a wood compensation strategy, and a plastic compensation strategy; The metal compensation strategy includes: by analyzing the brightness distribution of the image , using a local illumination correction algorithm, first calculating the average brightness of the local area , and then correcting the brightness of each pixel through a correction formula; The wood compensation strategy includes: filtering the image using a Gabor filter bank, and at the same time, performing dilation processing on the stain boundary using morphological operations; The plastic compensation strategy includes: adopting an image diffusion model based on partial differential equations, adaptively adjusting according to the image gradient, and enhancing the boundary of the stain area through the diffusion process.

Images