CN116703911A - LED lamp production quality detecting system - Google Patents
LED lamp production quality detecting system Download PDFInfo
- Publication number
- CN116703911A CN116703911A CN202310980121.6A CN202310980121A CN116703911A CN 116703911 A CN116703911 A CN 116703911A CN 202310980121 A CN202310980121 A CN 202310980121A CN 116703911 A CN116703911 A CN 116703911A
- Authority
- CN
- China
- Prior art keywords
- image
- connected domain
- merging
- range
- under
- 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.)
- Granted
Links
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 25
- 238000000034 method Methods 0.000 claims abstract description 54
- 238000001514 detection method Methods 0.000 claims abstract description 33
- 238000003708 edge detection Methods 0.000 claims abstract description 12
- 238000004891 communication Methods 0.000 claims description 27
- 230000008569 process Effects 0.000 claims description 18
- 238000001228 spectrum Methods 0.000 claims description 13
- 238000010606 normalization Methods 0.000 claims description 3
- 238000007689 inspection Methods 0.000 claims description 2
- 238000007670 refining Methods 0.000 claims description 2
- 230000007547 defect Effects 0.000 abstract description 17
- 238000012545 processing Methods 0.000 abstract description 3
- 230000008901 benefit Effects 0.000 abstract description 2
- 230000008859 change Effects 0.000 description 17
- 238000013528 artificial neural network Methods 0.000 description 7
- 238000009826 distribution Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000003313 weakening effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of image data processing, and provides an LED lamp production quality detection system, which comprises: acquiring a lamp panel image; the method comprises the steps of obtaining windows with different sizes, dividing a lamp panel image into a plurality of image blocks, obtaining reference weight values of the image blocks, and obtaining a target window range according to the reference weight values of the image blocks; acquiring a plurality of merging times, acquiring image areas according to the merging times under a target window range, acquiring connected domains according to edge detection results under different edge thresholds, acquiring key degrees and key information pixels of the connected domains according to the characteristics of the connected domains under the different edge thresholds, acquiring correlations among the image areas according to the key information pixels, and acquiring an optimal merging range; acquiring the preference degree of the target window range under the optimal combination range, and obtaining the optimal target range; and finishing quality detection according to the optimal target range. The invention ensures the advantage of larger difference between the defect information and the normal area information.
Description
Technical Field
The invention relates to the technical field of image data processing, in particular to a production quality detection system of an LED lamp.
Background
In recent years, LED (Light Emitting Diode) lamps have been widely used in the lighting industry. However, since there are certain drawbacks and variability in the production process of LED lamps, quality inspection is required to ensure that the product meets the standards. Conventional manual detection methods are time consuming, laborious and subject to subjective factors, and therefore there is a need to develop an efficient, accurate and automated quality detection method. The quality of the LED lamp is detected through the machine vision system, so that the quality of the produced LED lamp meets the standard, and a reliable illumination effect is provided. This will help to enhance product competitiveness and user satisfaction.
However, in the process of detecting the production quality of the LED lamp, the acquired image is blurred due to the influence of the resolution ratio of the acquired image and the influence of noise, and then larger errors can be generated when the characteristic information of the image is extracted later to serve as the LED lamp defect detection neural network. The fractal dimension information is various characteristics for representing the target object, the characteristic extraction can be better carried out, a typical algorithm is an MFDFA algorithm, but in the process of acquiring the fractal dimension information, the change of image information under different scales is required to be acquired, if the fractal dimension information is processed by using a traditional MFDFA algorithm, some key information is weakened due to the problem of scale selection, and then larger errors occur in the fractal dimension information of the follow-up acquired MFDFA. Therefore, in order to improve the detection result of the LED lamp defect detection neural network, the invention provides an MFDFA algorithm in a self-adaptive scale range, which acquires accurate fractal dimension information and performs accurate LED lamp production quality detection.
Disclosure of Invention
The invention provides a production quality detection system of an LED lamp, which aims to solve the problem that key information is weakened, and adopts the following technical scheme:
one embodiment of the invention provides a system for detecting the production quality of an LED lamp, which comprises the following modules:
the image acquisition module acquires a lamp panel image;
the target window range acquisition module is used for acquiring windows with different sizes, dividing a lamp panel image into a plurality of image blocks according to each window with different sizes, acquiring reference weight values of the image blocks according to gray values and the number of pixel points in the image blocks, acquiring the similarity of a local structure and an overall structure according to the reference weight values of the image blocks, and acquiring a target window range according to the similarity of the local structure and the overall structure;
the optimal merging range acquisition module acquires a plurality of merging times, acquires image areas according to the merging times under each target window range, uses edge detection for each image area, acquires connected domains according to edge detection results under different edge thresholds, and acquires the key degree of the connected domains according to the characteristics of the connected domains under the different edge thresholds; acquiring key information pixel points according to the key degree of the connected domain, acquiring a plurality of lines of curves in the image area according to the key information pixel points, acquiring the correlation between the image areas according to the lines of curves, and acquiring the optimal merging range according to the correlation between the image areas;
the optimal target window range obtaining module obtains the optimal degree of the target window range under the optimal combining range according to the number and gray values of the key information pixel points in the image block under the optimal combining range, and obtains the optimal target range according to the optimal degree of the target window range;
the quality detection module acquires a multi-fractal spectrum according to the optimal target range, and completes quality detection according to the multi-fractal spectrum.
Preferably, the method for dividing the lamp panel image into a plurality of image blocks according to each size window comprises the following steps:
the method comprises the steps of dividing a lamp panel image into a plurality of image blocks with window sizes, wherein gaps do not exist among the image blocks, and if the window exceeds the range of the lamp panel image, filling the exceeding part by using secondary linear interpolation, and dividing the lamp panel image into a plurality of image blocks.
Preferably, the method for obtaining the reference weight value of the image block according to the gray value and the number of the pixel points in the image block comprises the following steps:
each pixel point in the image block is marked as a reference pixel point, and the pixel points in eight adjacent areas of each reference pixel point are marked as neighborhood pixel points, if the absolute value of the gray value difference value between the reference pixel point and the corresponding neighborhood pixel point is smaller than a preset value, the neighborhood pixel point is considered as a similar pixel point of the reference pixel point;
in the method, in the process of the invention,gray value variance representing the nth image block at the ith window size, +.>Representing the number of similar pixels of all reference pixels in the nth image block at the ith window size, +.>Representing the total number of eight neighborhood pixel points of all reference pixel points in the nth image block under the ith window size, +.>Represents an exponential function based on natural constants, < ->Representing a linear normalization function, ++>Representing the reference weight value of the nth image block at the ith window size.
Preferably, the method for acquiring the image area according to the merging times under each target window range comprises the following steps:
and uniformly marking lines occupied by each image block under the current target window range as merging lines, merging the windows on the merging lines, merging the image blocks according to different merging times, wherein merging is carried out only on the same merging line, merging starts from the first image block of the merging line during merging, the merging refers to that two or more image blocks form an image area, if the same merging line merges, the rest image blocks cannot form a new image area, and then the rest image blocks are merged with the nearest image area on the same merging line, wherein the merging times are the number of the image block merging.
Preferably, the method for obtaining the connected domain according to the edge detection results under different edge thresholds and obtaining the key degree of the connected domain according to the characteristics of the connected domain under different edge thresholds comprises the following steps:
the connected domain under the detection of the initial edge threshold is marked as an initial connected domain, and connected domains at the same position under different edge thresholds are obtained according to the initial connected domain;
acquiring a communication angle difference between the communication domains according to the characteristic difference of the communication domains of the adjacent edge threshold values at the same position;
and acquiring the key degree of the initial connected domain according to the connected angle difference and the area of the connected domain.
Preferably, the method for obtaining the connected domain at the same position under different edge thresholds according to the initial connected domain comprises the following steps:
and marking the pixel points of the initial connected domain, and after the edge threshold value is changed, marking the connected domain with the marked pixel points in the connected domain acquired at the moment as a position connected domain, wherein the position connected domain and the initial connected domain are at the same position.
Preferably, the method for obtaining the communication angle difference between the communication domains according to the characteristic difference of the communication domains with adjacent edge thresholds at the same position includes:
the method comprises the steps of marking a connected domain at the same position of any edge threshold value with an initial connected domain as a standard connected domain, marking a connected domain at the same position of an adjacent threshold value of the edge threshold value corresponding to the standard connected domain as a continuous connected domain, extracting and refining the standard connected domain and the continuous connected domain by using a framework to obtain a framework line, connecting head and tail points of the framework line to obtain a straight line, marking the straight line as a standard line and a continuous line, and marking the absolute value of the angle difference value of the standard line and the continuous line as a connected angle difference;
if a plurality of continuous connected domains or standard connected domains exist, each continuous connected domain obtains a continuous line, each standard connected domain obtains a standard line, and the average value of absolute values of angle differences of the continuous line and the standard line is recorded as a connected angle difference.
Preferably, the method for obtaining the key degree of the initial connected domain according to the connected angle difference and the area of the connected domain comprises the following steps:
in the method, in the process of the invention,representing the communication angle difference between the communication domain at the same position with the s initial communication domain under the q-th edge threshold and the communication domain at the same position with the s initial communication domain under the q+1th edge threshold, and>representing the number of edge thresholds +.>Represents the area of the connected domain at the same position as the s-th initial connected domain under the q-th edge threshold,represents the area of the connected domain at the same position as the s-th initial connected domain under the q+1th edge threshold value, +.>Represents an exponential function based on natural constants, < ->Indicating the criticality of the s-th initial connected domain.
Preferably, the method for obtaining a plurality of rows of curves in the image area according to the key information pixel points comprises the following steps:
and marking the key information pixel points of each row in the image area from left to right, wherein the abscissa of the row curve is the serial number of the key information pixel points, and the ordinate of the row curve is the sum of the gray value of the key information pixel points and the gray value of the key information pixel points corresponding to the previous serial number.
Preferably, the method for obtaining the preference degree of the target window range under the optimal merging range according to the number and the gray value of the key information pixel points in the image block under the optimal merging range comprises the following steps:
in the method, in the process of the invention,representing the number proportion of key information pixels in the v-th image block under the optimal merging range,/>Representing the gray average value of key information pixels in the v-th image block under the optimal combination range,/>Representing the gray average value of all pixels in the v-th image block under the optimal merging range, ++>Representing the number of tiles in the j-th window range in the optimal merge range, +.>Indicating the preference degree of the target window range under the optimal merge range.
The beneficial effects of the invention are as follows: the invention provides an MFDFA algorithm in a self-adaptive scale range, which acquires accurate fractal dimension information as input data of an LED lamp defect detection neural network to perform accurate LED lamp production quality detection. According to the invention, the window range is obtained according to the distribution information of the LED lamp panels in the image to represent different scales, and the preference degree value of the scales is obtained according to the change of the distribution information under the different scales, so that the accurate fractal dimension information is obtained. The invention expects that the better scale is that the information of the image in the window range has great relativity, and the difference between the image area information with small relativity and the image area information with large relativity in the same window is great. The defect that some key information is weakened due to the problem of scale selection in the traditional MFDFA algorithm is avoided, the advantage of large difference between defect information and normal area information is ensured, and an accurate detection result can be obtained by the LED lamp defect detection neural network.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are only some embodiments of the invention, and that other drawings can be obtained according to these drawings without inventive faculty for a person skilled in the art.
Fig. 1 is a schematic flow chart of a system for detecting production quality of an LED lamp according to an embodiment of the present invention;
fig. 2 is a schematic diagram of image region merging.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, a flowchart of an LED lamp production quality detection system according to an embodiment of the present invention is shown, where the system includes: the device comprises an image acquisition module, a target window range acquisition module, an optimal combination range acquisition module, an optimal target window range acquisition module and a quality detection module.
The MiniLED lamp can use the lamp plate in the production process, and in the production process of using the lamp plate, an image acquisition device is arranged on the square of the lamp plate to acquire the lamp plate image in the production process, wherein the image acquisition device comprises a high-definition CCD camera, a light source, a bracket and a servo motor.
So far, the lamp panel image is acquired.
And the target window range acquisition module is used for acquiring accurate fractal dimension information by performing an adaptive MFDFA algorithm on the acquired lamp panel image. The method comprises the steps of analyzing the change of image information under different scales in the process of carrying out the MFDFA due to the influence of resolution and noise, and weakening some key information due to the problem of scale selection if the traditional MFDFA algorithm is used for processing, so that larger errors occur in fractal dimension information of the MFDFA acquired later. Therefore, in the embodiment, when fractal dimension information is used as input data of the LED lamp defect detection neural network, in order to adaptively acquire scale change information under different scales, a window range is acquired according to distribution information of an LED lamp panel in an image to represent different scales, and the LED lamp panel is often distributed repeatedly after the influence of resolution and noise is removed, namely, the correlation is larger; the present embodiment thus employs the entire image to be divided by the window range of the same size. The variation range of the window range is related to the acquisition of fractal dimension information, and the fractal dimension information is characterized by the similarity of a local structure and an overall structure, so that the window range selected by the scheme has a certain degree of similarity with the local structure and the overall structure, namely the similarity between the local structure and the overall structure of the corresponding window ranges does not change greatly in the dividing process.
Specifically, a side length is adopted for the selected windowAnd is an even number of squares, wherein the range of the side length is +.>Wherein A and B respectively represent the number of rows and columns of the lamp panel image, < >>Half of the minimum value representing the number of rows and columns of the panel image is rounded, and the similarity of the partial structure and the whole structure is analyzed in the window ranges.
For windows with different sizes, the lamp panel image is divided into image blocks with innumerable window sizes, wherein gaps do not exist among the image blocks, if the window exceeds the range of the lamp panel image, the exceeding part is filled by using secondary linear interpolation, and the lamp panel image is regarded as being composed of a plurality of image blocks.
In order to obtain the similarity of the local structure and the overall structure at the current window size, all image blocks need to be analyzed, different weights are applied to different image blocks, and the reference weight value of each image block is related to the complexity of the information in the image block. If the integral gray value change in the image block is not large and the pixel point distribution in the image block is disordered, the information contained in the image block is less, namely the corresponding reference weight value is smaller, each pixel point in the image block is marked as a reference pixel point, the pixel point in the eight adjacent areas of each reference pixel point is marked as a neighborhood pixel point, if the absolute value of the gray value difference value between the reference pixel point and the corresponding neighborhood pixel point is smaller than 10, the neighborhood pixel point is considered to be the similar pixel point of the reference pixel point, the number of the similar pixel points of all the reference pixel points in the image block is counted, wherein in the counting process, the pixel point can be counted repeatedly, if one pixel point is the similar pixel point of one reference pixel point, the pixel point can also be the similar pixel point of the other reference pixel point, and the reference weight value of the image block can be obtained according to the variance of the gray value in the image block and the similar pixel point quantity of the reference pixel point, and the reference weight formula is as follows:
in the method, in the process of the invention,gray value variance representing the nth image block at the ith window size, +.>Representing the number of similar pixels of all reference pixels in the nth image block at the ith window size, +.>Representing the total number of eight neighborhood pixel points of all reference pixel points in the nth image block under the ith window size, +.>Represents an exponential function based on natural constants, < ->Representing a linear normalization function, ++>Representing the reference weight value of the nth image block at the ith window size.
The structural similarity between each image block obtained by dividing under the current window and the lamp panel image is calculated, wherein the calculation method of the structural similarity is a technology known to the person skilled in the art, and the description is omitted again.
Obtaining the similarity of the local structure and the whole structure under the size of each window according to the obtained structural similarity of each image block and the lamp panel image and the reference weight value of each image block, wherein the formula is as follows:
in the method, in the process of the invention,reference weight value representing the nth image block at the ith window size, +.>Representing the structural similarity of the nth image block and the lamp panel image under the ith window size, +.>Representing the number of tiles at the ith window size,/v>Representing the similarity of partial and global structures at the size of the ith window, i.e., the ith windowSimilarity of local and global structures in the mouth region.
Setting a similarity thresholdWhen the similarity of the partial structure and the whole structure under the window range is larger than the threshold value, the window range at this time is taken as the target window range of the present embodiment, in the present embodiment +.>。
Thus, the target window range is acquired.
The best merging range obtaining module obtains a target window range according to the above, although the local and the whole of the windows have higher similarity, the target window ranges are not the window ranges expected by the embodiment, the best scale expected by the embodiment is that the information of the images in the window ranges has great correlation, and meanwhile, the difference between the image area information (irregular information, namely defect information) with small correlation and the image area information (regular information, namely normal information) with large correlation in the same window is great, so that the difference between the defect and the normal area in the fractal dimension information calculated under the corresponding window range is great, and the obtained target window ranges only meet the basic requirement of the fractal dimension information, but cannot meet the difference requirement of the image blocks after the window division.
When the target window range is acquired, the method only selects the calculation of the structural similarity, so that the acquired target window range is possibly too small, namely, one normal area is divided into a plurality of pixel blocks, and the pixel blocks basically only have the normal area and do not meet the difference requirement, so that the windows are required to be combined, and the image area information with large correlation is acquired according to the combination operation.
And analyzing the change of the image region information under different merging times for different target window ranges, and if the change is basically the same, the corresponding information in the region is an image region with larger correlation.
Specifically, for the line occupied by each image block under the current target window range, the line is collectively recorded as a merging line, for example, the current target window side length is 5, and then the merging line is represented by 5 pixel lines. And carrying out window merging on the merging parallel, obtaining merging times, wherein the merging times are 2 at least and half of the number of image blocks on one merging parallel at maximum, and rounding downwards if the merging times are decimal. The merging times are the number of image block merging, for example, the merging times are 2, which means that two image blocks are merged only in the same merging line, and merging is started from the first image block of the merging line during merging, so that the region after the image blocks are merged is taken as an image region, if the remaining image blocks cannot form a new image region during merging of the same merging line, the remaining image blocks are merged with the nearest image region on the same merging line, and the merging result is shown in fig. 2, wherein the image blocks corresponding to the same number in fig. 2 are merged into one image region.
And calculating the correlation between the image areas, wherein if the correlation between the image areas is larger, the image areas have stronger regularity.
In order to avoid the influence of resolution and noise of the lamp panel image, canny edge detection is performed on each image area, canny edge detection results under different edge thresholds are adopted in the embodiment, the Canny edge detection upper threshold is fixed to be 200, the lower threshold is initially 20, the threshold step length is 5, key information is acquired according to the edge detection results, correlation is acquired based on regularity of the key information, and the key information may exist under different thresholds. And (3) analyzing the image area after edge detection by using a connected domain to obtain a plurality of connected domains, marking the connected domain at the same position as the initial connected domain under any one edge threshold as a standard connected domain, marking the connected domain at the same position as a continuous connected domain, which is adjacent to the edge threshold corresponding to the standard connected domain, using a skeleton to extract and refine the standard connected domain and the continuous connected domain to obtain a skeleton line, connecting the head and tail points of the skeleton line to obtain a straight line as a standard line and a continuous line, marking the absolute value of the angle difference of the standard line and the continuous line as a connected angle difference, and marking each continuous line and marking the average value of the absolute value of the angle difference of the continuous line and the standard line as the connected angle difference if a plurality of continuous connected domains exist.
When analyzing the connected domain, taking the connected domain under the detection of the initial edge threshold value as an initial connected domain, taking the initial connected domain as a reference, marking the pixel point of the initial connected domain, and if the connected domain after the threshold value change has marked pixel points, considering the connected domain and the initial connected domain as the same position.
Obtaining the key degree of the connected domain according to the connected angle difference of different connected domains and the area of the connected domain under different thresholds, wherein the formula is as follows:
in the method, in the process of the invention,representing the communication angle difference between the communication domain at the same position with the s initial communication domain under the q-th edge threshold and the communication domain at the same position with the s initial communication domain under the q+1th edge threshold, and>representing the number of edge thresholds +.>Represents the area of the connected domain at the same position as the s-th initial connected domain under the q-th edge threshold,represents the area of the connected domain at the same position as the s-th initial connected domain under the q+1th edge threshold value, +.>Represents an exponential function based on natural constants, < ->Indicating the criticality of the s-th initial connected domain.
Representing the change of the angle of the straight line of the connected domain under different thresholds, wherein the change represents the regularity of the distribution of information in the graph, and the larger the corresponding value is, the smaller the key degree of the connected domain is; />The difference of the areas between the connected domain at the same position of the (q) th edge threshold and the connected domain at the same position of the(s) th initial connected domain and the connected domain at the same position of the (q+1) th edge threshold, if the difference of the connected domain areas is large, the connected domain is more discrete under the change of the corresponding different thresholds, and the key degree of the corresponding connected domain is smaller. The purpose of calculating the feature in this embodiment is to correct the change of the angle of the connected domain by the change of the area because there are some ghosts caused by the influence of the resolution in calculating the change of the angle of the connected domain under different thresholds so that the calculated key degree of the connected domain which is not key information is larger.
The key degree of each connected domain is obtained, the key degree of each connected domain is used as the key degree of a pixel point in the connected domain, a key degree threshold value is set, the key degree threshold value is 0.55 in the implementation, and if the key degree of the pixel point is larger than the key degree threshold value, the pixel point is indicated as the key information pixel point.
And analyzing the change rule of the key information in the image area according to the confirmed key information pixel points, so as to obtain the correlation between the image areas. If there is a strong correlation between the image areas, the change of the key information between the corresponding image areas is regular, so that for each line in the image area, a line curve is obtained, the abscissa of the line curve is the serial number of the key information pixel, the serial number is the label of the key information pixel of each line in the image area from left to right, the ordinate of the line curve is the sum of the gray value of the key information pixel and the gray value of the key information pixel corresponding to the previous serial number, for example, the serial numbers are 1,2,3,4,5, and the gray value of the key information pixel is 5,7,9,3,2, and the abscissa of the line curve is 1,2,3,4,5, and the ordinate is 5, 12, 21, 24, 26.
Acquiring a maximum value of an ordinate and a minimum value of the ordinate in each image area, acquiring a difference value of the maximum value and the minimum value of the ordinate, marking the difference value as a first difference value, acquiring a variance of the first difference value in all the image areas, combining any two image areas in all the image areas, calculating dtw distances in a one-to-one correspondence manner by each line curve, taking dtw distance average values of all the lines as distance values of any two image areas, and acquiring correlation between the image areas according to the variance of the first difference value and the dtw average values acquired by all the image areas, wherein the formula is as follows:
in the method, in the process of the invention,a first difference value representing an image area, +.>Representing the variance of the first difference values for all image areas,mean value of distance values representing all image areas between every two>Represents an exponential function based on natural constants, < ->Representing the correlation between the image areas.
If the corresponding range in the image areas is basically the same, the gray distribution characteristics of the two corresponding image areas are similar, namely the corresponding variation ranges are basically similar, and in order to analyze the interior of the image areas, the dtw distance between the row curves at the corresponding positions in the two image areas is taken as a reference weight, so that the correlation between the image areas is obtained.
Similarly, the correlation between the image areas under the rest merging times is obtained, the merging times with the largest correlation is selected from the obtained merging times, and a new merging range is obtained as the optimal merging range, so that the defect that the difference between the image area information with large correlation and the image area information with small correlation in the subsequent calculation is smaller due to the fact that the image is segmented by the target window range is avoided.
So far, the optimal merge scope size is obtained.
And the optimal target window range acquisition module is used for obtaining the optimal combination range of the target window ranges according to the steps, and calculating the optimal combination range of all the target window ranges to calculate the preference degree. Wherein the calculation of the degree of preference follows the principle that the difference between the image area information with small correlation and the image area information with large correlation is large. In the process of the correlation determined according to the above steps, image area information having a large correlation and image area information having a small correlation can be obtained.
Specifically, for all target window ranges under the optimal merging range, the number of image blocks under the window size is obtained, the image blocks at this time represent the merged image area, and the optimization degree of the target window range is obtained according to the number of the image blocks, the gray value of the key information pixel point in each image block and the gray value of all the pixel points in each image block, and the formula is as follows:
in the method, in the process of the invention,representing the number proportion of key information pixels in the v-th image block under the optimal merging range,/>Representing the gray average value of key information pixels in the v-th image block under the optimal combination range,/>Representing the gray average value of all pixels in the v-th image block under the optimal merging range, ++>Representing the number of tiles in the j-th window range in the optimal merge range, +.>Indicating the preference degree of the target window range under the optimal merge range.
And selecting a target window range corresponding to the maximum preference degree as an optimal target range.
So far, the optimal target range is obtained.
And the quality detection module is used for carrying out corresponding scale information of the MFDFA algorithm according to the optimal target range obtained by the steps. Under each scale, obtaining corresponding measurement fluctuation functions of different orders, carrying out logarithmic operation, carrying out line down fitting on the results obtained after logarithmic operation to obtain a generalized Hurst index, and further calculating a singular index and a corresponding singular spectrum to obtain a multi-fractal spectrum of the image. The process is a well-known technology, and will not be described in detail in this embodiment.
According to the acquired multi-fractal spectrum, acquiring corresponding fractal dimension information, wherein the acquisition comprises singular index maximum value, singular index minimum value, spectrum width, singular spectrum corresponding to the singular index maximum value, singular spectrum corresponding to the singular index minimum value, singular spectrum difference value between singular spectrum corresponding to the singular index maximum value and singular spectrum corresponding to the singular index minimum value. The characteristic values are used as input data of the LED lamp defect detection neural network, the LED lamp panels corresponding to the characteristic values of the samples are manually marked in a manual marking mode, different defect types are marked, for example, a foreign object defect area is marked as 1, a scratch defect area is marked as 2, a normal area is marked as 0 and the like, the adopted loss function is a cross entropy function, the LED lamp defect detection neural network is further trained, the obtained output is whether a certain defect exists or not, and then quality detection is completed.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.
Claims (10)
1. The LED lamp production quality detection system is characterized by comprising the following modules:
the image acquisition module acquires a lamp panel image;
the target window range acquisition module is used for acquiring windows with different sizes, dividing a lamp panel image into a plurality of image blocks according to each window with different sizes, acquiring reference weight values of the image blocks according to gray values and the number of pixel points in the image blocks, acquiring the similarity of a local structure and an overall structure according to the reference weight values of the image blocks, and acquiring a target window range according to the similarity of the local structure and the overall structure;
the optimal merging range acquisition module acquires a plurality of merging times, acquires image areas according to the merging times under each target window range, uses edge detection for each image area, acquires connected domains according to edge detection results under different edge thresholds, and acquires the key degree of the connected domains according to the characteristics of the connected domains under the different edge thresholds; acquiring key information pixel points according to the key degree of the connected domain, acquiring a plurality of lines of curves in the image area according to the key information pixel points, acquiring the correlation between the image areas according to the lines of curves, and acquiring the optimal merging range according to the correlation between the image areas;
the optimal target window range obtaining module obtains the optimal degree of the target window range under the optimal combining range according to the number and gray values of the key information pixel points in the image block under the optimal combining range, and obtains the optimal target range according to the optimal degree of the target window range;
the quality detection module acquires a multi-fractal spectrum according to the optimal target range, and completes quality detection according to the multi-fractal spectrum.
2. The LED lamp production quality inspection system of claim 1, wherein the method for dividing the lamp panel image into a plurality of image blocks according to each size window comprises:
the method comprises the steps of dividing a lamp panel image into a plurality of image blocks with window sizes, wherein gaps do not exist among the image blocks, and if the window exceeds the range of the lamp panel image, filling the exceeding part by using secondary linear interpolation, and dividing the lamp panel image into a plurality of image blocks.
3. The LED lamp production quality detection system according to claim 2, wherein the method for obtaining the reference weight value of the image block according to the gray value and the number of the pixels in the image block comprises:
each pixel point in the image block is marked as a reference pixel point, and the pixel points in eight adjacent areas of each reference pixel point are marked as neighborhood pixel points, if the absolute value of the gray value difference value between the reference pixel point and the corresponding neighborhood pixel point is smaller than a preset value, the neighborhood pixel point is considered as a similar pixel point of the reference pixel point;
in the method, in the process of the invention,gray value variance representing the nth image block at the ith window size, +.>Representing the number of similar pixels of all reference pixels in the nth image block at the ith window size, +.>Representing the total number of eight neighborhood pixel points of all reference pixel points in the nth image block under the ith window size, +.>Represents an exponential function with a base of a natural constant,representing a linear normalization function, ++>Representing the reference weight value of the nth image block at the ith window size.
4. The LED lamp production quality detection system of claim 1, wherein the method for obtaining the image area according to the merging times under each target window range comprises the following steps:
and uniformly marking lines occupied by each image block under the current target window range as merging lines, merging the windows on the merging lines, merging the image blocks according to different merging times, wherein merging is carried out only on the same merging line, merging starts from the first image block of the merging line during merging, the merging refers to that two or more image blocks form an image area, if the same merging line merges, the rest image blocks cannot form a new image area, and then the rest image blocks are merged with the nearest image area on the same merging line, wherein the merging times are the number of the image block merging.
5. The LED lamp production quality detection system according to claim 1, wherein the method for obtaining the connected domain from the edge detection results under different edge thresholds according to the features of the connected domain under different edge thresholds comprises the following steps:
the connected domain under the detection of the initial edge threshold is marked as an initial connected domain, and connected domains at the same position under different edge thresholds are obtained according to the initial connected domain;
acquiring a communication angle difference between the communication domains according to the characteristic difference of the communication domains of the adjacent edge threshold values at the same position;
and acquiring the key degree of the initial connected domain according to the connected angle difference and the area of the connected domain.
6. The system for detecting the production quality of an LED lamp according to claim 5, wherein the method for obtaining the connected domain at the same position under different edge thresholds according to the initial connected domain comprises the following steps:
and marking the pixel points of the initial connected domain, and after the edge threshold value is changed, marking the connected domain with the marked pixel points in the connected domain acquired at the moment as a position connected domain, wherein the position connected domain and the initial connected domain are at the same position.
7. The system for detecting the production quality of an LED lamp according to claim 5, wherein the method for obtaining the communication angle difference between the communication domains according to the characteristic difference of the communication domains at the same position of the adjacent edge threshold value comprises the following steps:
the method comprises the steps of marking a connected domain at the same position of any edge threshold value with an initial connected domain as a standard connected domain, marking a connected domain at the same position of an adjacent threshold value of the edge threshold value corresponding to the standard connected domain as a continuous connected domain, extracting and refining the standard connected domain and the continuous connected domain by using a framework to obtain a framework line, connecting head and tail points of the framework line to obtain a straight line, marking the straight line as a standard line and a continuous line, and marking the absolute value of the angle difference value of the standard line and the continuous line as a connected angle difference;
if a plurality of continuous connected domains or standard connected domains exist, each continuous connected domain obtains a continuous line, each standard connected domain obtains a standard line, and the average value of absolute values of angle differences of the continuous line and the standard line is recorded as a connected angle difference.
8. The system for detecting the production quality of an LED lamp according to claim 5, wherein the method for obtaining the key degree of the initial connected domain according to the connected angle difference and the area of the connected domain comprises the following steps:
in the method, in the process of the invention,representing the communication angle difference between the communication domain at the same position with the s initial communication domain under the q-th edge threshold and the communication domain at the same position with the s initial communication domain under the q+1th edge threshold, and>representing the number of edge thresholds +.>Represents the area of the connected domain at the same position as the s-th initial connected domain under the q-th edge threshold,/->Represents the area of the connected domain at the same position as the s-th initial connected domain under the q+1th edge threshold value, +.>Represents an exponential function based on natural constants, < ->Indicating the criticality of the s-th initial connected domain.
9. The LED lamp production quality detection system according to claim 1, wherein the method for obtaining a plurality of curves in the image area according to the key information pixels comprises:
and marking the key information pixel points of each row in the image area from left to right, wherein the abscissa of the row curve is the serial number of the key information pixel points, and the ordinate of the row curve is the sum of the gray value of the key information pixel points and the gray value of the key information pixel points corresponding to the previous serial number.
10. The system for detecting the production quality of an LED lamp according to claim 1, wherein the method for obtaining the preference degree of the target window range in the optimal combining range according to the number and the gray value of the key information pixels in the image block in the optimal combining range comprises the following steps:
in the method, in the process of the invention,representing the number proportion of key information pixels in the v-th image block under the optimal merging range,/>Representing the gray average value of key information pixels in the v-th image block under the optimal combination range,/>Representing the gray average value of all pixels in the v-th image block under the optimal merging range, ++>Representing the number of image blocks under the jth window range under the optimal merge range,indicating the preference degree of the target window range under the optimal merge range.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310980121.6A CN116703911B (en) | 2023-08-07 | 2023-08-07 | LED lamp production quality detecting system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310980121.6A CN116703911B (en) | 2023-08-07 | 2023-08-07 | LED lamp production quality detecting system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116703911A true CN116703911A (en) | 2023-09-05 |
| CN116703911B CN116703911B (en) | 2023-10-27 |
Family
ID=87843643
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310980121.6A Active CN116703911B (en) | 2023-08-07 | 2023-08-07 | LED lamp production quality detecting system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116703911B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117274820A (en) * | 2023-11-20 | 2023-12-22 | 深圳市天成测绘技术有限公司 | Map data acquisition method and system for mapping geographic information |
| CN117392114A (en) * | 2023-11-27 | 2024-01-12 | 湖南新域节能科技有限公司 | Betel nut grade detection method based on image processing |
| CN117788829A (en) * | 2024-02-27 | 2024-03-29 | 长春师范大学 | Image recognition system for invasive plant seed detection |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030163057A1 (en) * | 2002-02-22 | 2003-08-28 | Flick James T. | Method for diagnosing heart disease, predicting sudden death, and analyzing treatment response using multifractial analysis |
| WO2013115422A1 (en) * | 2012-02-02 | 2013-08-08 | 부산대학교 산학협력단 | Method for detecting objects using multifractal analysis of digital images |
| CN109961465A (en) * | 2017-12-21 | 2019-07-02 | 东莞东阳光科研发有限公司 | A kind of etched foil surface Tunnel-hole uniformity characterizing method based on multi-fractal |
| CN114137388A (en) * | 2021-11-25 | 2022-03-04 | 深圳市征阳电路科技有限公司 | Circuit board signal tracing detection system |
| CN115187548A (en) * | 2022-07-11 | 2022-10-14 | 常颖 | Mechanical part defect detection method based on artificial intelligence |
-
2023
- 2023-08-07 CN CN202310980121.6A patent/CN116703911B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030163057A1 (en) * | 2002-02-22 | 2003-08-28 | Flick James T. | Method for diagnosing heart disease, predicting sudden death, and analyzing treatment response using multifractial analysis |
| WO2013115422A1 (en) * | 2012-02-02 | 2013-08-08 | 부산대학교 산학협력단 | Method for detecting objects using multifractal analysis of digital images |
| CN109961465A (en) * | 2017-12-21 | 2019-07-02 | 东莞东阳光科研发有限公司 | A kind of etched foil surface Tunnel-hole uniformity characterizing method based on multi-fractal |
| CN114137388A (en) * | 2021-11-25 | 2022-03-04 | 深圳市征阳电路科技有限公司 | Circuit board signal tracing detection system |
| CN115187548A (en) * | 2022-07-11 | 2022-10-14 | 常颖 | Mechanical part defect detection method based on artificial intelligence |
Non-Patent Citations (2)
| Title |
|---|
| 简川霞: "手机玻璃屏表面缺陷机器视觉检测与分类方法研究", 《中国优秀博士学位论文全文数据库 信息科技辑》, no. 3, pages 138 - 31 * |
| 蒋爱平 等: "模拟X线头影片的多重分形分析", 《控制工程》, no. 06, pages 101 - 104 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117274820A (en) * | 2023-11-20 | 2023-12-22 | 深圳市天成测绘技术有限公司 | Map data acquisition method and system for mapping geographic information |
| CN117274820B (en) * | 2023-11-20 | 2024-03-08 | 深圳市天成测绘技术有限公司 | Map data acquisition method and system for mapping geographic information |
| CN117392114A (en) * | 2023-11-27 | 2024-01-12 | 湖南新域节能科技有限公司 | Betel nut grade detection method based on image processing |
| CN117392114B (en) * | 2023-11-27 | 2024-03-08 | 湖南新域节能科技有限公司 | Betel nut grade detection method based on image processing |
| CN117788829A (en) * | 2024-02-27 | 2024-03-29 | 长春师范大学 | Image recognition system for invasive plant seed detection |
| CN117788829B (en) * | 2024-02-27 | 2024-05-07 | 长春师范大学 | Image recognition system for invasive plant seed detection |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116703911B (en) | 2023-10-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116703911B (en) | LED lamp production quality detecting system | |
| CN111862064B (en) | Silver wire surface flaw identification method based on deep learning | |
| CN108960245B (en) | Tire mold character detection and recognition method, device, equipment and storage medium | |
| CN110148130B (en) | Method and device for detecting part defects | |
| CN110084241B (en) | Automatic ammeter reading method based on image recognition | |
| CN116205919B (en) | Hardware part production quality detection method and system based on artificial intelligence | |
| CN113554629A (en) | Strip steel red rust defect detection method based on artificial intelligence | |
| CN110490842B (en) | Strip steel surface defect detection method based on deep learning | |
| CN103593670A (en) | Copper sheet and strip surface defect detection method based on-line sequential extreme learning machine | |
| CN101059425A (en) | Method and device for identifying different variety green tea based on multiple spectrum image texture analysis | |
| CN108181316B (en) | Bamboo strip defect detection method based on machine vision | |
| CN116645367B (en) | Steel plate cutting quality detection method for high-end manufacturing | |
| CN115131354B (en) | Laboratory plastic film defect detection method based on optical means | |
| CN110334760B (en) | Optical component damage detection method and system based on RESUnet | |
| CN115861290B (en) | Skin-feel wood door surface defect detection method | |
| CN113935666B (en) | Building decoration wall tile abnormity evaluation method based on image processing | |
| CN113869300A (en) | Workpiece surface defect and character recognition method and system based on multi-vision fusion | |
| CN114926407A (en) | Steel surface defect detection system based on deep learning | |
| CN111753794A (en) | Fruit quality classification method and device, electronic equipment and readable storage medium | |
| CN114359235A (en) | Wood surface defect detection method based on improved YOLOv5l network | |
| CN111178405A (en) | Similar object identification method fusing multiple neural networks | |
| CN113610185B (en) | Wood color sorting method based on dominant hue identification | |
| CN115830514B (en) | Whole river reach surface flow velocity calculation method and system suitable for curved river channel | |
| CN112561875A (en) | Photovoltaic cell panel coarse grid detection method based on artificial intelligence | |
| CN116797575A (en) | Intelligent detection method for broken rice rate based on machine vision |
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 |