CN117522863B - Integrated box body quality detection method based on image features - Google Patents
Integrated box body quality detection method based on image features Download PDFInfo
- Publication number
- CN117522863B CN117522863B CN202311842781.4A CN202311842781A CN117522863B CN 117522863 B CN117522863 B CN 117522863B CN 202311842781 A CN202311842781 A CN 202311842781A CN 117522863 B CN117522863 B CN 117522863B
- Authority
- CN
- China
- Prior art keywords
- pixel
- edge
- integrated box
- image
- pixel points
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 39
- 238000001914 filtration Methods 0.000 claims abstract description 130
- 230000000694 effects Effects 0.000 claims abstract description 50
- 238000000034 method Methods 0.000 claims abstract description 50
- 230000007547 defect Effects 0.000 claims description 76
- 238000013528 artificial neural network Methods 0.000 claims description 12
- 230000000875 corresponding effect Effects 0.000 claims description 12
- 230000002596 correlated effect Effects 0.000 claims description 9
- 238000004891 communication Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 238000013527 convolutional neural network Methods 0.000 description 8
- 238000012549 training Methods 0.000 description 5
- 230000002950 deficient Effects 0.000 description 4
- 230000002708 enhancing effect Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000009825 accumulation Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000000739 chaotic effect Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000007847 structural defect Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 238000003466 welding Methods 0.000 description 1
Abstract
The invention relates to the technical field of image enhancement, in particular to an integrated box body quality detection method based on image characteristics. Firstly, checking an integrated box image by utilizing all connected domains and Gaussian filtering to carry out Gaussian filtering at least once to obtain a box enhanced image; in the process of determining the adjustment weight of each pixel point in the Gaussian filter kernel through each Gaussian filter, the adjustment weight of each pixel point is obtained according to the initial weight and the pixel participation degree of each pixel point; and further determining the detection result of the integrated box body quality. According to the invention, through determining the times of Gaussian filtering and the Gaussian filtering kernel, edge details of the integrated box body are not lost under the condition of efficiently filtering noise parts, the effect of filtering images is improved, and the quality detection accuracy of the integrated box body is further improved.
Description
Technical Field
The invention relates to the technical field of image enhancement, in particular to an integrated box body quality detection method based on image characteristics.
Background
The integrated housing is an important component of many devices, the quality of which is directly related to the stability and reliability of the device. If the integrated box body has defects, the durability of the integrated box body and the safety work of equipment are often affected, so that the integrated box body needs to be subjected to quality detection. In the prior art, the quality of the integrated box body can be detected by identifying the surface image of the integrated box body. Because the collected surface images of the integrated box body have noise influence, the edges of the integrated box body are difficult to accurately reflect, and the collected surface images of the integrated box body need to be subjected to enhanced filtering, so that the quality of the integrated box body is analyzed.
In the prior art, the collected surface image of the integrated box body can be subjected to enhanced filtering through Gaussian filtering, noise reduction of the surface image of the integrated box body can be realized through Gaussian filtering, and image contrast is enhanced, but the Gaussian filtering effect is influenced by Gaussian filtering times and Gaussian filtering kernels. The influence of the Gaussian filtering times on the Gaussian filtering effect is shown as follows: if the Gaussian filtering times are too small, the filtered image is insufficient to achieve the noise reduction effect; if the number of Gaussian filtering times is too large, the filtered image is excessively smoothed and edge details of the excessive image are lost. The effect of the Gaussian filter check on the Gaussian filter effect is mainly reflected in that: on the one hand, when the weight of the pixel points in the Gaussian filter kernel is unreasonable, the phenomenon of blurring of the image can be caused; on the other hand, when the size of the determined gaussian filter kernel is too small, only a local area of the integrated box image can be covered, part of pixels may not be sufficiently smoothed, so that accuracy of a filtering result is affected, and when the size of the determined gaussian filter kernel is too large, details and features of the integrated box image are lost, so that accuracy of the filtering result is affected. In the prior art, the Gaussian filtering times and the Gaussian filtering kernel are usually determined according to experience, so that edge details of the integrated box body are not lost under the condition that the existing Gaussian filtering is difficult to realize efficient noise filtering, and further the filtering image effect is poor, and the quality detection effect of the integrated box body is inaccurate.
Disclosure of Invention
In order to solve the technical problems that edge details of an integrated box body are not lost under the condition that the existing Gaussian filter is difficult to realize high-efficiency noise filtering, the filtering image effect is poor, and the quality detection effect of the integrated box body is inaccurate, the invention aims to provide an integrated box body quality detection method based on image characteristics, and the adopted technical scheme is as follows:
an integrated case quality detection method based on image features, the method comprising the steps of:
acquiring an integrated box image;
acquiring all connected domains and Gaussian filter kernels of the integrated box image; checking the integrated box body image by utilizing all the connected domains and Gaussian filtering to carry out Gaussian filtering at least once, and obtaining a box body enhanced image;
the step of carrying out Gaussian filtering on the integrated box image at least once to obtain a box enhanced image comprises the following steps: in the process of determining the adjustment weight of each pixel point in the Gaussian filter kernel through each Gaussian filter, acquiring the initial weight of each pixel point in the Gaussian filter kernel; in the preset surrounding neighborhood of each pixel point, taking the pixel point corresponding to the edge of the connected domain as an edge pixel point; acquiring the pixel participation degree of each pixel point in a preset surrounding neighborhood of each pixel point according to the number of the edge pixel points and the linear shape characteristics of the edge pixel points; acquiring the adjustment weight of each pixel point according to the initial weight and the pixel participation degree of each pixel point;
and determining an integrated box body quality detection result according to the box body enhanced image.
Further, the method for acquiring the pixel participation degree comprises the following steps:
taking the total number of edge pixel points in the preset surrounding neighborhood as the total number of edge pixel points of the pixel points;
taking the total number of the pixel points in the preset surrounding neighborhood as the total number of the pixel points;
acquiring a first edge possible value of the pixel points according to the total number of the edge pixel points and the total number of the pixel points; the total number of the edge pixel points and the first edge possible value are in positive correlation; the total number of the pixel points and the possible value of the first edge are in negative correlation;
performing straight line fitting on all the edge pixel points to obtain all fitting straight lines;
calculating the average value of the straight line fitting rate of all fitting straight lines to obtain a second edge possible value of the pixel point;
acquiring the pixel participation degree of the pixel point according to the first edge possible value and the second edge possible value; the first edge likelihood value and the pixel engagement degree are in positive correlation; the second edge likelihood value and the pixel engagement are positively correlated.
Further, the method for acquiring the adjustment weight comprises the following steps:
acquiring an adjustment weight of the pixel point according to the initial weight of the pixel point and the pixel participation degree;
the initial weight and the adjusted weight are positively correlated; the pixel engagement and the adjustment weight are positively correlated.
Further, the step of obtaining the gaussian filter kernel includes:
acquiring the distribution characteristic angle of each connected domain according to the distribution characteristics of each connected domain in the integrated box image;
acquiring the filter kernel size of the integrated box image according to the distribution characteristic angles of all the connected domains and the sizes of all the connected domains;
and determining a Gaussian filter kernel of the integrated box body image according to the filter kernel size of the integrated box body image.
Further, the obtaining formula of the filter kernel size includes:
the method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>A filter kernel size for the integrated box image; />Is the total number of connected domains; />Is->Maximum span values of the connected domains; />Is->Minimum span values of the connected domains; />The average value of the distribution characteristic angles of all the connected domains is obtained; />Is->The distribution characteristic angles of the connected domains; />Is a normalization function; />Rounding up the symbol; />Is an absolute value sign.
Further, the method for acquiring the distribution characteristic angle comprises the following steps:
acquiring the mass centers of all the connected domains;
taking any one connected domain as a current connected domain, and acquiring Euclidean distances between the current connected domain and centroids of all other connected domains;
the communication domain corresponding to the minimum Euclidean distance is used as a reference communication domain of the current communication domain;
constructing a straight line through the centroids of the current connected domain and the reference connected domain, and taking the straight line as a characteristic straight line of the current connected domain;
and taking the included angle between the characteristic straight line and the set direction as the distribution characteristic angle of the current connected domain.
Further, the step of performing at least one gaussian filtering on the integrated box image further includes:
after each Gaussian filtering, determining target pixel points in a preset neighborhood range of each pixel point according to the gray values of the pixel points in the preset neighborhood range of each pixel point, acquiring edge filtering effect values of each pixel point according to the gray values of the pixel points in the preset neighborhood range of each pixel point and the target pixel points, and acquiring the current filtering times of each pixel point; and until the edge filtering effect values or the filtering times of all the pixel points meet a preset iteration termination condition, all the pixel points do not perform the next Gaussian filtering.
Further, the formula for obtaining the edge filtering effect value includes:
the method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>Is->The edge filtering effect values of the pixel points; />In the +.>The average value of gray values of all the pixel points in the preset neighborhood range of the pixel points;in->In the preset neighborhood range of each pixel point, the average value of gray values of all the pixel points in all the connected domains; />In the +.>The total number of all target pixel points in the preset neighborhood range of each pixel point; />In->The total number of all the pixel points in all the connected domains in the preset neighborhood range of the pixel points; />Is a normalization function.
Further, the method for acquiring the connected domain includes:
extracting all edges in the integrated box image by using a watershed algorithm; taking the end-to-end closed edges as target edges; and taking the area surrounded by the target edge as a connected area.
Further, the method for acquiring the integrated box body quality detection result comprises the following steps:
acquiring all defect areas in the box enhanced image based on a neural network; determining defect indexes of each defect area by extracting the area of each defect area in the box enhanced image, wherein the area of each defect area is positively correlated with the defect indexes;
calculating the accumulated sum of the defect indexes of each defect area, and determining defect parameters; and determining the quality detection result of the integrated box body according to the defect parameters.
The invention has the following beneficial effects:
the invention needs to carry out enhanced filtering on the integrated box body image, and in order to realize noise reduction and enhance the visibility of the edge of the integrated box body, the traditional Gaussian filtering needs to be improved. In order to make the pixel point in the background area lower, the pixel point participation degree is higher as the pixel point in the edge part is positioned. Because in the surrounding area of the pixel points in the background area and the noise area, the number of edge pixel points around the pixel points is less, the edge pixel points are difficult to perform straight line fitting, in the surrounding area of the pixel points in the edge area, the number of edge pixel points around the pixel points is more, the edge pixel points can perform straight line fitting, and the pixel participation degree of each pixel point is obtained by combining the number of the edge pixel points and the straight line shape characteristics, wherein the pixel participation degree reflects the possibility that the pixel points are in the edge area; on the basis of determining the initial weight of each pixel point in the Gaussian filter kernel through the Gaussian function, acquiring the adjustment weight of each pixel point through the pixel participation degree; so as to achieve the purpose of increasing the visibility of the edges of the integrated box body. And further obtaining a box body enhanced image and determining an integrated box body quality detection result. The invention has more accurate edge expression effect by enhancing the image, and finally effectively improves the accuracy of the quality inspection result of the integrated box.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions and advantages of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are only some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flowchart of an integrated box quality detection method based on image features according to an embodiment of the present invention.
Detailed Description
In order to further describe the technical means and effects adopted by the present invention to achieve the preset purpose, the following detailed description refers to the specific implementation, structure, characteristics and effects of an integrated box body quality detection method based on image characteristics according to the present invention with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" means that the embodiments are not necessarily the same. Furthermore, the particular features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
An embodiment of an integrated box body quality detection method based on image features:
the following specifically describes a specific scheme of the integrated box body quality detection method based on image features.
Referring to fig. 1, a flowchart of an integrated box quality detection method based on image features according to an embodiment of the invention is shown, the method includes the following steps:
and S1, acquiring an integrated box image.
The integrated box is an important component of a plurality of devices, the quality of the integrated box influences the safety work of the devices, and in order to analyze the quality of the integrated box, an integrated box image needs to be acquired first for subsequent filtering enhancement of the integrated box image, so that the defects of the integrated box are analyzed more accurately.
Specifically, in the embodiment of the invention, in order to avoid the influence of direct sunlight, overcast and rainy days and the like on the imaging effect, a proper acquisition place is selected and a proper shooting distance and angle are kept for the camera so as to ensure that a clear original surface image of the integrated box body can be acquired. And enhancing the contrast of the original surface image by using histogram equalization, and acquiring the integrated box image. The embodiment of the invention adopts histogram equalization to enhance contrast, and an operator can set a noise reduction method according to actual conditions.
It should be noted that, in order to facilitate the operation, all index data involved in the operation in the embodiment of the present invention is subjected to data preprocessing, so as to cancel the dimension effect. The specific means for removing the dimension influence is a technical means well known to those skilled in the art, and is not limited herein.
S2, acquiring all connected domains and Gaussian filter kernels of the integrated box image; and checking the integrated box body image by utilizing all the connected domains and Gaussian filtering to carry out Gaussian filtering at least once, and obtaining a box body enhanced image.
Because the Gaussian filter kernel is too small, only a local area of the integrated box image can be covered, partial pixel points can not be sufficiently smoothed, and the accuracy of a filtering result is affected; however, the details and characteristics of the integrated box image are lost due to the overlarge Gaussian filter kernel, and the accuracy of a filtering result is affected, so that the size of the Gaussian filter kernel needs to be determined, and the Gaussian filter kernel with the better size is selected. The structure of the edges of the normal integrated box body is regular, and the regularity of the areas formed by the edges is high; the structure of the edge of the abnormal integrated box body is disordered, and the regularity of the area formed by the edge is low; therefore, each connected domain of the integrated box image is acquired, and the connected domain can reflect the area surrounded by the edges. And determining a Gaussian filter kernel with a proper size by utilizing all the connected domains, so that the Gaussian filter kernel is integrated with each pixel point in the box image according to the Gaussian filter kernel for multiple times to acquire a box enhanced image, and the box enhanced image result is accurate and the calculated amount in the filtering process is not excessive.
Preferably, in one embodiment of the present invention, the method for acquiring the gaussian filter kernel includes:
because the defects of the integrated box body are mainly structural deformation, the structural deformation is mainly caused by poor welding, and the structural irregularities of the edge area are shown in the integrated box body image; when the structural quality of the integrated box body is higher, the structure of the edge of the integrated box body is more regular, and the regularity of the area formed by the edge is higher. Therefore, the more chaotic the distribution of the connected domains is, the less the distribution is close to the structural rule distribution characteristics of the integrated box body, so that the distribution characteristic angles of all the connected domains are required to be obtained according to the distribution characteristics of all the connected domains in the integrated box body image, and the distribution characteristic angles reflect the distribution conditions of the connected domains. The more chaotic the distribution of the connected domain is, the smaller the reference around the pixel point is, and the smaller the Gaussian filter kernel is; in order to effectively reflect and observe the structural characteristics of the integrated box body, the larger the size of the connected domain is, the larger the Gaussian filter kernel is required to be amplified. And determining the size of the filter kernel through the size of the connected domain and the distribution disorder degree of the connected domain, and further determining the Gaussian filter kernel of the integrated box body image according to the size of the filter kernel of the integrated box body image. The Gaussian filter kernel has a better size so as to ensure the accuracy of a filtering result and the calculated amount is not excessively large.
Preferably, in order to analyze the regularity of the region constituted by the upper edge of the integrated tank, it is first necessary to obtain the connected domain constituted by the edge. In one embodiment of the present invention, the method for obtaining the connected domain includes:
extracting all edges in the integrated box body image by using a watershed algorithm; taking the end-to-end closed edges as target edges; and taking the area surrounded by the target edge as a connected area. The connected domain can reflect the area surrounded by the edges for subsequent analysis of the structural features of the integrated box.
Preferably, as the regularity of the region formed by the edges is higher, if the distribution of the connected domains is more disordered, the scene features are distributed more rarely to the structural rules of the integrated box body, and the distribution condition of the connected domains is reflected by the distribution feature angles of the connected domains. In one embodiment of the present invention, the method for obtaining the distribution characteristic angle includes:
acquiring the mass centers of all the connected domains;
taking any connected domain as a current connected domain, and acquiring Euclidean distances between centroids of the current connected domain and all other connected domains;
the minimum Euclidean distance is used as a reference connected domain of the current connected domain;
constructing a straight line through the mass centers of the current connected domain and the reference connected domain, and taking the straight line as a characteristic straight line of the current connected domain;
and taking the included angle between the characteristic straight line and the set direction as the distribution characteristic angle of the current connected domain, wherein the distribution characteristic angle reflects the distribution condition of the connected domain.
In the embodiment of the present invention, when a two-dimensional coordinate system is established on the integrated box image, in the present embodiment, on the basis of ensuring that one or a group of sides of the integrated box are horizontally or vertically distributed in the integrated box image, for example, when the integrated box image is a front view of the integrated box placed standing, sides corresponding to heights of the integrated box are vertically distributed in the integrated box image, at this time, a lower left corner of the integrated box image is taken as an origin of coordinates, a horizontal right direction is taken as an X-axis direction, and a vertical upward direction is taken as a Y-axis direction, so that the two-dimensional coordinate system on the integrated box image is established, and the set direction is parallel to the X-axis direction. Of course, as another embodiment, the image coordinate system may be directly used as a reference, and the two-dimensional coordinate system may be established by integrating the box image, and the direction may be set to be parallel to the X-axis direction.
Preferably, in one embodiment of the present invention, the method for obtaining the filter kernel size includes:
and acquiring the size of a filtering kernel of the integrated box body image through the size of the connected domain and the distribution disorder degree of the connected domain so as to determine a Gaussian filtering kernel with a proper size later. The filter kernel size obtaining formula in one embodiment of the present invention includes:
the method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>The size of a filtering kernel of the integrated box body image; />Is the total number of connected domains; />Is->Maximum span values of the connected domains; />Is->Minimum span values of the connected domains; />The average value of the distribution characteristic angles of all the connected domains; />Is->Distribution characteristic angles of the connected domains;is a normalization function; />Rounding up the symbol; />Is an absolute value sign. In one embodiment of the invention, any two edge pixel points connected with the connected domain determine each line segment, the line segment passing through the center point of the connected domain is taken as a target line segment, the maximum value in the lengths of all the target line segments is taken as a maximum span value, and the minimum value in the lengths of all the target line segments is taken as a minimum span value. Of course, as another embodiment, the maximum span value may be the length of the smallest circumscribed rectangle of the connected domain, and in this case, the minimum span value is the width of the smallest circumscribed rectangle of the connected domain.
In the filter kernel size formula, the filter kernel size is calculatedAs->By reflecting the distribution disorder of connected domain and the area pair of connected domain +.>And (5) adjusting. Because of the high regularity of the integrated box image, < > and so on>Through the distribution characteristic angles of all connected domainsThe average value of the integrated box body image can reflect the overall distribution angle of the integrated box body image; />The deviation of the distribution angle and the whole distribution angle of the connected domains is reflected, and when the deviation is larger, the distribution among the corresponding connected domains is disordered, so that the lower the reference of the connected domains is, the lower the weight is. By->Preventing denominator from being 0. In order to effectively reflect the structural features of the observation integrated box body, < > the following>The larger the filter kernel size should be, the higher the weight. />Reflecting the length of the smallest circumscribed rectangle of the connected domain, < >>The larger the filter kernel size is, the larger the filter kernel size is. Since the length of the gaussian filter kernel is often a positive integer, the filter kernel size needs to be rounded up. The size of the filtering core comprehensively reflects the distribution disorder and the size of the connected domain, and can reflect the structural rule characteristics of the integrated box structure, so that the accuracy of the filtering result is ensured later and the calculated amount is not excessive.
Specifically, in order to make the gaussian filter kernel have a preferable size, the gaussian filter kernel of the integrated box image is determined according to the filter kernel size of the integrated box image. In the embodiment of the invention, the size of the Gaussian filter kernel is*/>，/>For the size of the filtering kernel, all connected domains and Gaussian filtering cores are further utilized to check the integrated box body image to carry out Gaussian filtering for a plurality of times, and the box is obtainedThe volume enhances the image.
Step S3, in the process of determining the adjustment weight of each pixel point in the Gaussian filter kernel through each Gaussian filter, obtaining the initial weight of each pixel point in the Gaussian filter kernel; acquiring the pixel participation degree of each pixel point in a preset surrounding neighborhood of each pixel point according to the number of the edge pixel points and the linear shape characteristics of the edge pixel points; and acquiring the adjustment weight of each pixel point according to the initial weight and the pixel participation degree of each pixel point.
The invention needs to carry out enhanced filtering on the integrated box body image, and in order to realize noise reduction and enhance the visibility of the edge of the integrated box body, the traditional Gaussian filtering needs to be improved. In the process of determining the adjustment weight of each pixel point in the Gaussian filter kernel by each Gaussian filter, on the basis of determining the initial weight of each pixel point in the Gaussian filter kernel, the participation degree of the pixel points is lower for the pixel points which are positioned in the background area and the noise area; the more the pixel points are positioned at the edge part, the higher the participation degree of the pixel points is. Acquiring pixel participation degree of each pixel point in a preset surrounding neighborhood of each pixel point according to the number of the edge pixel points and the linear shape characteristics of the edge pixel points, wherein the pixel participation degree reflects the possibility that the pixel point is in an edge area, and acquiring adjustment weights of the pixel points through the pixel participation degree; the initial weight is adjusted by the probability that the pixel points are in the edge area on the basis of retaining the Gaussian filter kernel weight, so that the Gaussian filter achieves the purpose of reducing noise and increasing the visibility of the edge of the integrated box body.
It should be noted that, the gaussian filtering is a technical means well known to those skilled in the art, and is not described herein, and the order of the gaussian filtering is usually from left to right and from top to bottom. When processing the integrated box image, firstly, the integrated box image starts from the pixel at the upper left corner, then moves to the right, and after processing the pixels of one row, moves downwards and continues to process the pixels of the next row. This order is because the gaussian filter is a transversal convolution kernel that requires each pixel to be processed in line order.
Specifically, in order to adjust the weight in the gaussian filter core based on the conventional gaussian filter, the original initial weight is first obtained. And using the Gaussian function to take the weight of each pixel point in the Gaussian filter kernel as the initial weight of each pixel point.
Preferably, in one embodiment of the present invention, a method for acquiring a pixel participation degree includes:
in the preset surrounding neighborhood of the pixel point, the pixel point corresponding to the edge of the connected domain is used as an edge pixel point;
the total number of edge pixel points in the preset surrounding neighborhood is used as the total number of edge pixel points of the pixel points;
the total number of the pixel points in the surrounding neighborhood is preset and is used as the total number of the pixel points;
acquiring a first edge possible value of the pixel points according to the total number of the edge pixel points and the total number of the pixel points; the total number of the edge pixel points and the possible value of the first edge are in positive correlation; the total number of pixel points and the possible value of the first edge are in negative correlation;
performing straight line fitting on all edge pixel points to obtain all fitting straight lines;
calculating the average value of the straight line fitting rate of all fitting straight lines to obtain a second edge possible value of the pixel point;
acquiring the pixel participation degree of the pixel point according to the first edge possible value and the second edge possible value; the first edge likelihood value and the pixel participation degree are in positive correlation; the second edge likelihood value and the pixel engagement exhibit a positive correlation.
In order to achieve noise reduction and enhance the visibility of the edge of the integrated box, taking the possibility that a pixel is in an edge area into consideration, the pixel participation degree of the pixel is obtained, and an obtaining formula of the pixel participation degree in one embodiment of the invention comprises the following steps:
the method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>Is->Pixel participation of each pixel point; />Is->In the preset surrounding neighborhood of each pixel point, the total number of edge pixel points; />Is->In the preset surrounding neighborhood of each pixel point, the total number of the pixel points;is->Second edge possible values of the individual pixel points; />Is->A first edge possible value of the pixel point. In one embodiment of the invention, a preset surrounding neighborhood is constructed by taking a pixel point as a center point; the center of the preset surrounding neighborhood is a center point, the preset surrounding neighborhood is b x b, b is the size of the filtering kernel, and an implementer can set the filtering kernel according to implementation scenes.
In the pixel participation formula, as the main characteristic of the integration box body is regularity, the edge of the integration box body accords with the straight line shape,the probability that the pixel point is in the edge area is reflected through the linear fitting rate of the pixel points around the pixel point, and the greater the probability, the greater the pixel participation degree. />Reflect->The larger the duty ratio is, the greater the likelihood that the pixel is in the edge region, and the greater the pixel participation. Since there are often few edge pixels around the pixel in the background region and the noise region and the edge pixels are difficult to fit straight lines, the pixel participation is lower when the pixel is the background region pixel or the noise pixel. The pixel participation degree integrates the duty ratio of the edge pixel points around the pixel points and the linear fitting rate of the edge pixel points around the pixel points, and more comprehensively reflects the possibility that the pixel points are in the edge area, and the greater the possibility that the pixel points are the edge pixel points.
Preferably, in one embodiment of the present invention, the method for obtaining the adjustment weight includes:
acquiring an adjustment weight of the pixel point according to the initial weight of the pixel point and the pixel participation degree;
the initial weight and the adjustment weight are in positive correlation; the pixel engagement and adjustment weights are positively correlated.
On the basis of reserving Gaussian filter kernel weights, initial weights are adjusted through the possibility that pixel points are located in edge areas, and adjustment weights are obtained. In one embodiment of the present invention, the obtaining formula for adjusting the weight includes:
the method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>Is->Adjusting weights of the pixel points; />Is->Initial weights of the individual pixels; />Is->Pixel participation of each pixel point.
In the weight adjustment formula, the initial weight is adjusted through the pixel participation degree, and as different pixel points have different pixel participation degrees, for partial noise pixel points or background pixel points, the pixel participation degree is lower when the filtering output is calculated, so that the lower the corresponding adjustment weight is in the subsequent iterative calculation filtering result, and the influence of the noise points and the background points on the filtering output is eliminated; for the pixel points in the edge area, the pixel participation degree is higher when the filtering output is calculated, so that the corresponding adjustment weight is higher in the subsequent iterative calculation filtering result, and the influence of the pixel points with the edge characteristics on the filtering output is amplified. The weight is adjusted so that Gaussian filtering achieves the purpose of reducing noise and increasing the visibility of the edges of the integrated box body.
Step S4, after Gaussian filtering is carried out each time, obtaining an edge filtering effect value of each pixel point according to the gray value of the pixel point in the preset neighborhood range of each pixel point, and obtaining the current filtering times of each pixel point; and until the edge filtering effect values or the filtering times of all the pixel points meet the preset iteration termination condition, not performing Gaussian filtering on all the pixel points next time.
Too many filtering times can cause excessive smoothing of the image and loss of too many edge details, and too few filtering times can be difficult to filter noise. In the process of filtering noise and enhancing the visibility of edges, determining proper filtering times, after each Gaussian filtering, acquiring the edge filtering effect value of each pixel point according to the gray value of the pixel point in the preset neighborhood range of each pixel point, and acquiring the current filtering times of each pixel point; the purpose of the edge filtering effect value and the filtering times is to limit the iteration times of the filtering, and when the edge filtering effect value or the filtering times after the filtering for a plurality of times meet the preset iteration termination condition, the iteration is stopped; because different pixel points have different edge characteristics and different filtering times, the integrated box image is subjected to filtering under the self-adaptive times, the noise reduction is realized, the edge expression effect of the integrated box is enhanced, and the image after the filtering enhancement is obtained.
Preferably, in one embodiment of the present invention, the method for obtaining the edge filtering effect value includes:
the method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>Is->Edge filtering effect values of the pixel points; />In the +.>The average value of gray values of all pixel points in a preset neighborhood range of each pixel point; />In->In a preset neighborhood range of each pixel point, the average value of gray values of all the pixel points in all the connected domains; />In the +.>The total number of all target pixel points in the preset neighborhood range of each pixel point; />In->Preset neighborhood of each pixel pointIn the range, the total number of all pixel points in all connected domains; />Is a normalization function. In one embodiment of the invention, a pixel point is taken as a central pixel point, and a preset neighborhood range is constructed; the center of the preset neighborhood range is a central pixel point, the size of the preset neighborhood range is c, c is the size of the filter kernel, and an implementer can set the preset neighborhood range according to implementation scenes.
In the edge filtering effect value formula,reflecting the gray value average value around the pixel point; />The gray value mean value of the connected domain around the pixel point is reflected, and the gray values of the integrated box are different from the gray values of the background area because the area formed by the edges of the integrated box are regularly distributed; />Reflected in->Around each pixel point, integrating the difference between the gray value of the box body and the background area, wherein the larger the difference is, the better the filtering effect is, and the larger the edge filtering effect value is; />The closer the ratio is to 1, the closer the filtered result is to the result after division by the Ojin method, the closer the filtering effect is to the real division result, and the better the filtering effect is, the larger the value of the edge filtering effect is; the edge filtering effect value comprehensively reflects the filtering effect of the pixel point in the edge region.
Preferably, in order to analyze the segmentation result that is closer to the true in terms of filtering effect, the target pixel point is first acquired. In one embodiment of the present invention, the method for obtaining the target pixel point includes:
image segmentation is carried out on a preset neighborhood range by using an Ojin threshold method, and a foreground image of the preset neighborhood range is obtained; acquiring all foreground pixel points in a foreground image; and taking all foreground pixel points as all target pixel points.
Because different pixel points have different edge characteristics and different edge filtering effect values, the integrated box image is filtered under the self-adaption times, the visibility of the edge of the integrated box is enhanced while noise reduction is realized, and the filtered and enhanced image is obtained. In one embodiment of the present invention, a method for acquiring a box enhanced image includes:
after each Gaussian filtering, if the edge filtering effect value or the filtering times of the pixel point meet the preset iteration termination condition, the pixel point terminates the iteration; if the edge filtering effect value or the filtering times of the pixel point do not meet the preset iteration termination condition, continuing iteration of the pixel point; and obtaining the box enhanced image until the edge filtering effect values or the filtering times of all the pixel points meet the preset iteration termination condition. In one embodiment of the present invention, the preset iteration termination condition is that the edge filter effect value is greater than the preset edge effect value or the iteration number is greater than the preset iteration number. Wherein, the preset edge effect value is 0.8, the preset iteration number is 2, and the implementer can set according to the implementation scene.
And S5, determining an integrated box body quality detection result according to the box body enhanced image.
The box enhanced image is noise-reduced, meanwhile, the visibility of the edge of the integrated box is enhanced, and the detection result of the quality of the integrated box is determined. The enhanced image has a more accurate edge expression effect, so that the accuracy of defect area detection and the final quality inspection result are more accurate.
Preferably, in one embodiment of the present invention, the box enhanced image enhances the visibility of the edges of the integrated box while reducing noise, and improves the accuracy of detecting the edges of the defective areas, so that the neural network can converge more efficiently and can recognize more efficiently when recognizing the defects, thereby improving the accuracy of the detection result of the integrated box body. The method for acquiring the detection result of the integrated box body quality comprises the following steps:
acquiring all defect areas in the box enhanced image based on a neural network; determining defect indexes of each defect area by extracting the area of each defect area in the box enhanced image, wherein the area of the defect area is positively correlated with the defect indexes;
calculating the accumulation sum of defect indexes of each defect area, and determining defect parameters; and determining the quality detection result of the integrated box body according to the defect parameters. In one embodiment of the invention, all defective areas in the box enhanced image are acquired through a CNN neural network.
It should be noted that, the CNN neural network is a technical means well known to those skilled in the art, and is not described herein in detail, but only a brief process of determining all defect areas in the enhanced image of the box by using the CNN neural network in one embodiment of the present invention is described briefly:
and taking a plurality of integrated box images with structural defects as a reference data set, training a CNN neural network by using the reference data set, and marking the defect areas of the reference data set. Dividing the marked reference data set into a training set and a verification set according to a preset proportion, inputting the reference data set into the CNN neural network for training, adopting a binary cross entropy loss function as a loss function, and adopting a gradient descent method until the loss function converges to obtain the CNN neural network after training; and inputting the box enhanced image into a CNN neural network after training, and marking the defective areas by the CNN neural network to obtain all the defective areas in the box enhanced image. In the embodiment of the invention, the preset ratio is 7:3, a step of; the optimizer selects SGD to increase convergence speed and performance of the model, and an implementer can set the SGD according to implementation scenes.
Specifically, determining defect indexes of each defect area by counting the area of each defect area, setting the defect grade of the corresponding defect area as a first defect when the area of the defect area is larger than a first defect area threshold, and setting the first defect as a first defect index; when the area of the defect area is larger than the second defect area threshold and is not larger than the first defect area threshold, setting the defect grade of the corresponding defect area as a second defect, and setting the second defect as a second defect index; when the area of the defect area is not greater than a third defect area threshold, setting the defect grade of the corresponding defect area as a third defect, and setting the third defect as a third defect index; in this embodiment, the value of the first defect area threshold is set to be 50, the value of the second defect area threshold is set to be 20, the value of the first defect index is set to be 10, the value of the second defect index is set to be 4, the value of the third defect index is set to be 1, and the operator can set the values according to the implementation scene.
After determining defect indexes of areas of all defect areas in the extracted image, calculating accumulation of the defect indexes and obtaining defect parameters, and determining the quality detection grade of the integrated box body according to the defect parameters. When the defect parameter is smaller than the first set parameter, judging that the quality detection grade of the integrated box body is excellent; when the defect parameter is not smaller than the first set parameter and smaller than the second set parameter, judging that the quality detection grade of the integrated box body is good; when the defect parameter is not smaller than the second setting parameter and smaller than the third setting parameter, judging that the quality detection grade of the integrated box body is qualified; and when the defect parameter is not smaller than the third set parameter, judging that the quality detection grade of the integrated box body is unqualified, and needing to be reworked. In this embodiment, the value of the first setting parameter is set to be 1, the value of the second setting parameter is set to be 5, and the value of the third setting parameter is set to be 10, so that the operator can set the setting according to the implementation scenario. Thus, the detection result of the integrated box body quality is determined.
In summary, the embodiment of the invention provides an integrated box body quality detection method based on image characteristics, which comprises the steps of firstly acquiring all connected domains and Gaussian filter kernels of an integrated box body image; checking the integrated box body image by utilizing all connected domains and Gaussian filtering to carry out Gaussian filtering for a plurality of times, and obtaining a box body enhanced image; in the process of determining the adjustment weight of each pixel point in the Gaussian filter kernel through each Gaussian filter, the adjustment weight of each pixel point is obtained according to the initial weight and the pixel participation degree of each pixel point; until the edge filtering effect values or the filtering times of all the pixel points meet the preset iteration termination condition, all the pixel points do not carry out the next Gaussian filtering; and further determining the detection result of the integrated box body quality. According to the invention, through determining the times of Gaussian filtering and the Gaussian filtering kernel, edge details of the integrated box body are not lost under the condition of efficiently filtering noise parts, the effect of filtering images is improved, and the quality detection accuracy of the integrated box body is further improved.
It should be noted that: the sequence of the embodiments of the present invention is only for description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments.
Claims (7)
1. An integrated box body quality detection method based on image features, which is characterized by comprising the following steps:
acquiring an integrated box image;
acquiring all connected domains and Gaussian filter kernels of the integrated box image; checking the integrated box body image by utilizing all the connected domains and Gaussian filtering to carry out Gaussian filtering at least once, and obtaining a box body enhanced image;
the step of carrying out Gaussian filtering on the integrated box image at least once to obtain a box enhanced image comprises the following steps: in the process of determining the adjustment weight of each pixel point in the Gaussian filter kernel through each Gaussian filter, acquiring the initial weight of each pixel point in the Gaussian filter kernel; in the preset surrounding neighborhood of each pixel point, taking the pixel point corresponding to the edge of the connected domain as an edge pixel point; acquiring the pixel participation degree of each pixel point in a preset surrounding neighborhood of each pixel point according to the number of the edge pixel points and the linear shape characteristics of the edge pixel points; acquiring the adjustment weight of each pixel point according to the initial weight and the pixel participation degree of each pixel point;
determining an integrated box body quality detection result according to the box body enhanced image;
the method for acquiring the pixel participation degree comprises the following steps:
taking the total number of edge pixel points in the preset surrounding neighborhood as the total number of edge pixel points of the pixel points;
taking the total number of the pixel points in the preset surrounding neighborhood as the total number of the pixel points;
acquiring a first edge possible value of the pixel points according to the total number of the edge pixel points and the total number of the pixel points; the total number of the edge pixel points and the first edge possible value are in positive correlation; the total number of the pixel points and the possible value of the first edge are in negative correlation;
performing straight line fitting on all the edge pixel points to obtain all fitting straight lines;
calculating the average value of the straight line fitting rate of all fitting straight lines to obtain a second edge possible value of the pixel point;
acquiring the pixel participation degree of the pixel point according to the first edge possible value and the second edge possible value; the first edge likelihood value and the pixel engagement degree are in positive correlation; the second edge likelihood value and the pixel engagement degree are in positive correlation;
the step of performing at least one gaussian filtering on the integrated box image further comprises:
after each Gaussian filtering, determining target pixel points in a preset neighborhood range of each pixel point according to the gray values of the pixel points in the preset neighborhood range of each pixel point, acquiring edge filtering effect values of each pixel point according to the gray values of the pixel points in the preset neighborhood range of each pixel point and the target pixel points, and acquiring the current filtering times of each pixel point; until the edge filtering effect values or the filtering times of all the pixel points meet a preset iteration termination condition, all the pixel points do not carry out Gaussian filtering for the next time;
the formula for obtaining the edge filtering effect value comprises the following steps:
the method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>Is->The edge filtering effect values of the pixel points; />In the +.>The average value of gray values of all the pixel points in the preset neighborhood range of the pixel points; />In->In the preset neighborhood range of each pixel point, the average value of gray values of all the pixel points in all the connected domains; />In the +.>The total number of all target pixel points in the preset neighborhood range of each pixel point; />In->The total number of all the pixel points in all the connected domains in the preset neighborhood range of the pixel points; />Is a normalization function.
2. The method for detecting the quality of an integrated box body based on image features according to claim 1, wherein the method for acquiring the adjustment weight comprises the following steps:
acquiring an adjustment weight of the pixel point according to the initial weight of the pixel point and the pixel participation degree;
the initial weight and the adjusted weight are positively correlated; the pixel engagement and the adjustment weight are positively correlated.
3. The method for detecting the quality of an integrated box body based on image features according to claim 1, wherein the step of obtaining the gaussian filter kernel comprises:
acquiring the distribution characteristic angle of each connected domain according to the distribution characteristics of each connected domain in the integrated box image;
acquiring the filter kernel size of the integrated box image according to the distribution characteristic angles of all the connected domains and the sizes of all the connected domains;
and determining a Gaussian filter kernel of the integrated box body image according to the filter kernel size of the integrated box body image.
4. A method for detecting the quality of an integrated box based on image features according to claim 3, wherein the formula for obtaining the size of the filter kernel comprises:
the method comprises the steps of carrying out a first treatment on the surface of the Wherein (1)>A filter kernel size for the integrated box image; />Is the total number of connected domains; />Is->Maximum span values of the connected domains; />Is->Minimum span values of the connected domains; />The average value of the distribution characteristic angles of all the connected domains is obtained; />Is->The distribution characteristic angles of the connected domains; />Is a normalization function; />Rounding up the symbol; />Is an absolute value sign.
5. The method for detecting the quality of an integrated box body based on image features according to claim 3, wherein the method for acquiring the distribution feature angle comprises the following steps:
acquiring the mass centers of all the connected domains;
taking any one connected domain as a current connected domain, and acquiring Euclidean distances between the current connected domain and centroids of all other connected domains;
the communication domain corresponding to the minimum Euclidean distance is used as a reference communication domain of the current communication domain;
constructing a straight line through the centroids of the current connected domain and the reference connected domain, and taking the straight line as a characteristic straight line of the current connected domain;
and taking the included angle between the characteristic straight line and the set direction as the distribution characteristic angle of the current connected domain.
6. The method for detecting the quality of an integrated tank based on image features as claimed in claim 1, wherein the method for acquiring the connected domain comprises:
extracting all edges in the integrated box image by using a watershed algorithm; taking the end-to-end closed edges as target edges; and taking the area surrounded by the target edge as a connected area.
7. The method for detecting the quality of an integrated tank based on image features according to claim 1, wherein the method for acquiring the quality detection result of the integrated tank comprises the following steps:
acquiring all defect areas in the box enhanced image based on a neural network; determining defect indexes of each defect area by extracting the area of each defect area in the box enhanced image, wherein the area of each defect area is positively correlated with the defect indexes;
calculating the accumulated sum of the defect indexes of each defect area, and determining defect parameters; and determining the quality detection result of the integrated box body according to the defect parameters.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311842781.4A CN117522863B (en) | 2023-12-29 | 2023-12-29 | Integrated box body quality detection method based on image features |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311842781.4A CN117522863B (en) | 2023-12-29 | 2023-12-29 | Integrated box body quality detection method based on image features |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117522863A CN117522863A (en) | 2024-02-06 |
| CN117522863B true CN117522863B (en) | 2024-03-29 |
Family
ID=89745970
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311842781.4A Active CN117522863B (en) | 2023-12-29 | 2023-12-29 | Integrated box body quality detection method based on image features |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117522863B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008107002A1 (en) * | 2007-03-05 | 2008-09-12 | Fotonation Vision Limited | Face searching and detection in a digital image acquisition device |
| WO2017092431A1 (en) * | 2015-12-01 | 2017-06-08 | 乐视控股（北京）有限公司 | Human hand detection method and device based on skin colour |
| CN107358618A (en) * | 2017-07-20 | 2017-11-17 | 深圳市唯特视科技有限公司 | A kind of method at faint edge in detection noise image |
| CN108109155A (en) * | 2017-11-28 | 2018-06-01 | 东北林业大学 | A kind of automatic threshold edge detection method based on improvement Canny |
| WO2020248439A1 (en) * | 2019-06-11 | 2020-12-17 | 江苏农林职业技术学院 | Crown cap surface defect online inspection method employing image processing |
| CN114549529A (en) * | 2022-04-26 | 2022-05-27 | 武汉福旺家包装有限公司 | Carton indentation quality detection method and system based on computer vision |
| WO2022205525A1 (en) * | 2021-04-01 | 2022-10-06 | 江苏科技大学 | Binocular vision-based autonomous underwater vehicle recycling guidance false light source removal method |
| WO2023097931A1 (en) * | 2021-12-03 | 2023-06-08 | 江苏航天大为科技股份有限公司 | Hough transform-based license plate tilt correction method |
| CN116309570A (en) * | 2023-05-18 | 2023-06-23 | 山东亮马新材料科技有限公司 | Titanium alloy bar quality detection method and system |
| CN116452594A (en) * | 2023-06-19 | 2023-07-18 | 安徽百胜电子系统集成有限责任公司 | Visualized monitoring and early warning method and system for power transmission line state |
| CN116777907A (en) * | 2023-08-18 | 2023-09-19 | 深圳市福山自动化科技有限公司 | Sheet metal part quality detection method |
| CN116934763A (en) * | 2023-09-19 | 2023-10-24 | 阿普塔恒煜(威海)医疗器材有限公司 | Medical rubber plug defect detection method based on visual characteristics |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060274962A1 (en) * | 2005-06-03 | 2006-12-07 | Yi-Jen Chiu | Systems and methods for improved Gaussian noise filtering |
| EP2677464B1 (en) * | 2012-05-16 | 2018-05-02 | IMEC vzw | Feature detection in numeric data |
| US10430943B2 (en) * | 2016-10-07 | 2019-10-01 | Sony Corporation | Automated nuclei area/number estimation for IHC image analysis |
-
2023
- 2023-12-29 CN CN202311842781.4A patent/CN117522863B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008107002A1 (en) * | 2007-03-05 | 2008-09-12 | Fotonation Vision Limited | Face searching and detection in a digital image acquisition device |
| WO2017092431A1 (en) * | 2015-12-01 | 2017-06-08 | 乐视控股（北京）有限公司 | Human hand detection method and device based on skin colour |
| CN107358618A (en) * | 2017-07-20 | 2017-11-17 | 深圳市唯特视科技有限公司 | A kind of method at faint edge in detection noise image |
| CN108109155A (en) * | 2017-11-28 | 2018-06-01 | 东北林业大学 | A kind of automatic threshold edge detection method based on improvement Canny |
| WO2020248439A1 (en) * | 2019-06-11 | 2020-12-17 | 江苏农林职业技术学院 | Crown cap surface defect online inspection method employing image processing |
| WO2022205525A1 (en) * | 2021-04-01 | 2022-10-06 | 江苏科技大学 | Binocular vision-based autonomous underwater vehicle recycling guidance false light source removal method |
| WO2023097931A1 (en) * | 2021-12-03 | 2023-06-08 | 江苏航天大为科技股份有限公司 | Hough transform-based license plate tilt correction method |
| CN114549529A (en) * | 2022-04-26 | 2022-05-27 | 武汉福旺家包装有限公司 | Carton indentation quality detection method and system based on computer vision |
| CN116309570A (en) * | 2023-05-18 | 2023-06-23 | 山东亮马新材料科技有限公司 | Titanium alloy bar quality detection method and system |
| CN116452594A (en) * | 2023-06-19 | 2023-07-18 | 安徽百胜电子系统集成有限责任公司 | Visualized monitoring and early warning method and system for power transmission line state |
| CN116777907A (en) * | 2023-08-18 | 2023-09-19 | 深圳市福山自动化科技有限公司 | Sheet metal part quality detection method |
| CN116934763A (en) * | 2023-09-19 | 2023-10-24 | 阿普塔恒煜(威海)医疗器材有限公司 | Medical rubber plug defect detection method based on visual characteristics |
Non-Patent Citations (4)
| Title |
|---|
| 基于k-邻域前景像素分析的划痕提取方法;李家宏;陈斌;;计算机应用;20171220(S2);正文全文 * |
| 基于边缘检测的高斯噪声滤波算法;吴翰;江巨浪;;安庆师范大学学报(自然科学版);20180315(01);正文全文 * |
| 改进高斯滤波算法的舰船尾迹图像无损去噪技术;李俊峰;;舰船科学技术;20200923(18);正文全文 * |
| 高斯噪声下的人脸检测算法研究;朱强;徐颖;王壮才;;数字技术与应用;20200525(05);正文全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117522863A (en) | 2024-02-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107808378B (en) | Method for detecting potential defects of complex-structure casting based on vertical longitudinal and transverse line profile features | |
| CN107507173B (en) | No-reference definition evaluation method and system for full-slice image | |
| WO2022110804A1 (en) | Image noise measurement method and device based on local statistical information | |
| CN113538433A (en) | Mechanical casting defect detection method and system based on artificial intelligence | |
| WO2021109697A1 (en) | Character segmentation method and apparatus, and computer-readable storage medium | |
| CN109492642B (en) | License plate recognition method, license plate recognition device, computer equipment and storage medium | |
| CN107742307A (en) | Based on the transmission line galloping feature extraction and parameters analysis method for improving frame difference method | |
| CN115330645B (en) | Welding image enhancement method | |
| CN106530281A (en) | Edge feature-based unmanned aerial vehicle image blur judgment method and system | |
| CN115797473B (en) | Concrete forming evaluation method for civil engineering | |
| CN114926407A (en) | Steel surface defect detection system based on deep learning | |
| CN105389793A (en) | Method for automatically identifying trend and width of fracture in object surface | |
| CN113313107A (en) | Intelligent detection and identification method for multiple types of diseases on cable surface of cable-stayed bridge | |
| CN109712116B (en) | Fault identification method for power transmission line and accessories thereof | |
| CN109954854B (en) | Method and device for monitoring breakout of crystallizer of continuous casting machine, storage medium and electronic terminal | |
| CN117522863B (en) | Integrated box body quality detection method based on image features | |
| CN113436102A (en) | Method for detecting surface defects of machined part | |
| CN111428538B (en) | Lane line extraction method, device and equipment | |
| CN113436120A (en) | Image fuzzy value identification method and device | |
| CN111415365A (en) | Image detection method and device | |
| CN114004812A (en) | Threaded hole detection method and system adopting guide filtering and neural network model | |
| CN109272540B (en) | SFR automatic extraction and analysis method of image of graphic card | |
| CN117495863B (en) | Vision-assisted triode packaging quality detection method | |
| CN107491746B (en) | Face pre-screening method based on large gradient pixel analysis | |
| CN112085683A (en) | Depth map reliability detection method in significance detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |