CN104359918A - Method for detecting surface defects of speaker cone - Google Patents

Abstract

The invention relates to a method for detecting surface defects of a speaker cone. The method comprises the following steps: acquiring a surface image of the speaker cone, and segmenting a to-be-detected annular area; performing gradient detection on the to-be-detected annular area; and finally, performing consistency detection on the to-be-detected area. By utilizing the non-contact digital image detection technology, the defect detection efficiency of the speaker cone is improved, the method can be used for automatically detecting the surface appearance defects of the speaker cone, the product quality can be further improved, the production cost is reduced, and the method has important practical engineering significances.

Description

A kind of detection method of surface flaw of diffuser

Technical field

The invention belongs to field of machine vision, be specifically related to a kind of detection method of surface flaw of diffuser.

Background technology

In loudspeaker, can produce sharp pounding in magnetic field after voice coil loudspeaker voice coil connects sound signal, voice coil loudspeaker voice coil is fixed on cone, and the vibrations of voice coil loudspeaker voice coil drive the vibrations of cone simultaneously, and cone vibrations force air to shake, thus sound.Manufacturer is in the process manufacturing cone, and the reason cone surface due to technique there will be the surface imperfection such as cone slurry mark, cone lumps, cone wood chip, cone white point, affects outward appearance and the result of use of cone.In actual industrial production process, due to the restriction of technical conditions, cone surface defects detection still rests in the level of artificial visually examine, and labor strength is large, and production efficiency is low.And manual detection method is by the impact of the factors such as human eye vision sensitivity, individual subjective judgement, usually accurately and reliably cannot catch defect information, produce a large amount of undetected and flase drops by being easy to like this.If careless manipulation in the detection, also secondary damage can be caused to cone.Therefore work out a kind of open defect detection method being suitable for cone surface, realize Aulomatizeted Detect, can improve the quality of products further, reduce production cost, there is important practical meaning in engineering.

Summary of the invention

The object of the invention is the deficiency in order to overcome artificial visually examine's method, utilizing non-contact digital image detecting technique to improve the efficiency of diffuser defects detection, proposing a kind of detection method of surface flaw of diffuser.Concrete steps:

Step (1) gathers the surface image of diffuser, is partitioned into annular region to be detected;

The cone image F (size is M × N) collected; Due to the particular requirement of subsequent technique, only need to detect the annular region of cone; So need to carry out region segmentation to the original image collected, obtain annular cone region to be detected:

A () carries out Threshold segmentation, η by following formula to the cone image collected ₁elect 80 as, obtain prospect cone region:

$F(i,j)=\left\{\begin{array}{c}0,\mathrm{ifF}(i,j)<{\mathrm{\&eta;}}_1\\ F(i,j),\mathrm{else}\end{array}\right.$

B () utilizes horizontal and vertical to project and positions the center of circle of cone, finally segmentation obtains the annular cone region needing to detect:

By following formula, horizontal projection is done to cone image:

$H\left(i\right)=\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}F(i,j)$

Then i is increased to M from 1, when occurring that H (i) is greater than threshold value η ₂time (η ₂elect 5000 as), current i is the coboundary of cone; Then i is increased, when occurring that H (i) is less than threshold value η ₂time, current i is the lower limb of cone;

By following formula, vertical projection is done to cone image:

$V\left(j\right)=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}F(i,j)$

Then j is increased to N from 1, when occurring that V (j) is greater than threshold value η ₂time, current j is the left hand edge of cone; Then j is increased, when occurring that V (j) is less than threshold value η ₂time, current j is the right hand edge of cone;

By horizontal and vertical projection, the edge, upper and lower, left and right of cone is demarcated, thus obtain the central coordinate of circle of cone.Then according to cone size and camera resolution, select the interior external radius of annular region, annular region to be detected is split;

Step (2) carries out gradient detection to the annular region to be detected split;

The defect on cone surface mainly contains cone lumps, cone wood chip, cone white point and cone slurry mark.Gradient detection algorithm is adopted to detect cone lumps, cone wood chip, cone white-spot defects; First adopt the gaussian filtering kernel function of 5 × 5 to carry out gaussian filtering to cone annular region to be detected, remove the impact of noise;

Grad f ' (i, j) corresponding to each point is calculated as follows to filtered annular region f:

$f^{\&prime;}(i,j)=\frac{\left|f_x^{\&prime;}(i,j)\right|+\left|f_y^{\&prime;}(i,j)\right|}{2}$

Wherein horizontal and vertical gradient is calculated as follows:

f′ _x(i,j)＝f(i+1,j)-f(i-1,j)

f′ _y(i,j)＝f(i,j+1)-f(i,j-1)

Then threshold process is done to gradient image f ', η ₃elect 8 as, remove the impact of noise further:

$f^{\&prime;}(i,j)=\left\{\begin{array}{c}0,f^{\&prime;}(i,j)<{\mathrm{\&eta;}}_3\\ f^{\&prime;}(i,j),\mathrm{else}\end{array}\right.$

The number of the interior non-zero points of gradient image f ' after last statistical threshold process, if not zero number is greater than threshold alpha, α elects 15 as, then judge this cone existing defects; Otherwise consistency detection is carried out to this cone region;

Step (3) is treated surveyed area and is made consistency detection;

Due to the restriction of technique, there will be cone slurry mark defect at the fringe region of cone; Due to the defect that cone slurry mark defect is on texture, the slurry mark defect of gradient detection algorithm to tiny like this texture is difficult to detect; Ring-type segmentation is carried out to the annular of cone region to be detected, because cone slurry mark defect only appears in the outer ledge region of cone, so only do consistency detection to outermost two-layer annular region.Ground floor annular region is positioned at outermost, and second layer annular region is positioned at the inner side of ground floor, and the radial width of annular region is determined by actual cone size and camera resolution.Each segmented areas of ground floor annular region is designated as each segmented areas of second layer annular region is designated as calculate the color average of each piecemeal in each layer annular region with color variance

Every one deck annular region is judged as follows:

$R_i^1=\left\{\begin{array}{cc}1,& \mathrm{if}(M_i^1/(\underset{i}{\mathrm{\&Sigma;}}{M}_i^1/K)>{\mathrm{\&beta;}}_1)\mathrm{or}(V_i^1/(\underset{i}{\mathrm{\&Sigma;}}{V}_i^1/K)>{\mathrm{\&beta;}}_2)\\ 0,& \mathrm{else}\end{array}\right.$

$R_i^2=\left\{\begin{array}{cc}1,& \mathrm{if}(M_i^2/(\underset{i}{\mathrm{\&Sigma;}}{M}_i^2/K)>{\mathrm{\&beta;}}_1)\mathrm{or}(V_i^2/(\underset{i}{\mathrm{\&Sigma;}}{V}_i^2/K)>{\mathrm{\&beta;}}_2)\\ 0,& \mathrm{else}\end{array}\right.$

Here K represents the piecemeal number in every layer of annular region, threshold value beta ₁, β ₂elect 1.3 as;

If the mark of all piecemeals is 0, then judge this cone not existing defects, otherwise judge this cone existing defects.

The present invention is by gradient and consistency detection, and the adaptive surface appearance defects to diffuser detects.

Beneficial effect of the present invention: the inventive method utilizes non-contact digital image detecting technique to improve the efficiency of diffuser defects detection, proposes a kind of detection method of surface flaw of diffuser.The inventive method is detected by gradient and the adaptive surface imperfection to diffuser of region consistency detection method detects, and can well overcome the deficiency of artificial visually examine's detection method, reduce production cost, have important practical meaning in engineering.

Accompanying drawing explanation

Fig. 1 a is original image;

Fig. 1 b is region to be detected;

Fig. 2 is that region consistency detects partitioning scheme figure

Embodiment

Below in conjunction with accompanying drawing, the invention will be further described, the present invention includes following steps:

Step (1) gathers the surface image of diffuser, is partitioned into annular region to be detected;

The cone image F (size is M × N) collected, as shown in Figure 1a, due to the particular requirement of subsequent technique, only need to detect (as shown in Figure 1 b) the annular region of cone, so need to carry out region segmentation to the original image collected, obtain annular cone region to be detected:

A () carries out threshold value η by following formula to the cone image collected ₁segmentation obtains prospect cone region:

$F(i,j)=\left\{\begin{array}{c}0,\mathrm{ifF}(i,j)<{\mathrm{\&eta;}}_1\\ F(i,j),\mathrm{else}\end{array}\right.$

B () utilizes horizontal and vertical to project and positions the center of circle of cone, finally segmentation obtains the annular cone region needing to detect:

By following formula, horizontal projection is done to cone image:

$H\left(i\right)=\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}F(i,j)$

Then i is increased to M from 1, when occurring that H (i) is greater than threshold value η ₂time, current i is the coboundary of cone; Then i is increased, when occurring that H (i) is less than threshold value η ₂time, current i is the lower limb of cone;

By following formula, vertical projection is done to cone image:

$V\left(j\right)=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}F(i,j)$

Then j is increased to N from 1, when occurring that V (j) is greater than threshold value η ₂time, current j is the left hand edge of cone; Then j is increased, when occurring that V (j) is less than threshold value η ₂time, current j is the right hand edge of cone;

By horizontal and vertical projection, the edge, upper and lower, left and right of cone is demarcated, thus obtain the central coordinate of circle of cone.Then according to cone size and camera resolution, select the interior external radius of annular region, annular region to be detected is split;

Step (2) carries out gradient detection to the annular region to be detected split;

The defect on cone surface mainly contains cone lumps, cone wood chip, cone white point and cone slurry mark.Gradient detection algorithm is adopted to detect cone lumps, cone wood chip, cone white-spot defects; First adopt the gaussian filtering kernel function of 5 × 5 to carry out gaussian filtering to cone annular region to be detected, remove the impact of noise;

Grad f ' (i, j) corresponding to each point is calculated as follows to filtered annular region f:

$f^{\&prime;}(i,j)=\frac{\left|f_x^{\&prime;}(i,j)\right|+\left|f_y^{\&prime;}(i,j)\right|}{2}$

Wherein horizontal and vertical gradient is calculated as follows:

f′ _x(i,j)＝f(i+1,j)-f(i-1,j)

f′ _y(i,j)＝f(i,j+1)-f(i,j-1)

Then threshold value η is done to gradient image f ' ₃process, remove the impact of noise further:

$f^{\&prime;}(i,j)=\left\{\begin{array}{c}0,f^{\&prime;}(i,j)<{\mathrm{\&eta;}}_3\\ f^{\&prime;}(i,j),\mathrm{else}\end{array}\right.$

The number of the interior non-zero points of gradient image f ' after last statistical threshold process, if not zero number is greater than threshold alpha, then judges this cone existing defects; Otherwise consistency detection is carried out to this cone region;

Step (3) is treated surveyed area and is made consistency detection;

Due to the restriction of technique, there will be cone slurry mark defect at the fringe region of cone; Due to the defect that cone slurry mark defect is on texture, the slurry mark defect of gradient detection algorithm to tiny like this texture is difficult to detect; Split by the annular of mode to the cone region to be detected of Fig. 2, each segmented areas of ground floor annular region is designated as each segmented areas of second layer annular region is designated as calculate the color average of each piecemeal in each layer annular region with color variance

Every one deck annular region is judged as follows:

$R_i^1=\left\{\begin{array}{cc}1,& \mathrm{if}(M_i^1/(\underset{i}{\mathrm{\&Sigma;}}{M}_i^1/K)>{\mathrm{\&beta;}}_1)\mathrm{or}(V_i^1/(\underset{i}{\mathrm{\&Sigma;}}{V}_i^1/K)>{\mathrm{\&beta;}}_2)\\ 0,& \mathrm{else}\end{array}\right.$

$R_i^2=\left\{\begin{array}{cc}1,& \mathrm{if}(M_i^2/(\underset{i}{\mathrm{\&Sigma;}}{M}_i^2/K)>{\mathrm{\&beta;}}_1)\mathrm{or}(V_i^2/(\underset{i}{\mathrm{\&Sigma;}}{V}_i^2/K)>{\mathrm{\&beta;}}_2)\\ 0,& \mathrm{else}\end{array}\right.$

Here K represents the piecemeal number in every layer of annular region, β ₁, β ₂be respectively average and variance threshold values;

If the mark in all piecemeals is 0, then judge this cone not existing defects, otherwise judge this cone existing defects.

Claims (1)

1. a detection method of surface flaw for diffuser, is characterized in that the concrete steps of the method are:

Step (1) gathers the surface image of diffuser, is partitioned into annular region to be detected, specifically:

If the cone image F collected, size is M × N;

A () carries out threshold value η by following formula to the cone image collected ₁segmentation, thus obtain prospect cone region:

$F(i,j)=\left\{\begin{array}{cc}0,& \mathrm{if\; F}(i,j)<{\mathrm{\&eta;}}_1\\ F(i,j),& \mathrm{else}\end{array}\right.$

B () utilizes horizontal and vertical to project and positions the center of circle of cone, finally segmentation obtains the annular cone region needing to detect:

By following formula, horizontal projection is done to cone image:

$H\left(i\right)=\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}F(i,j)$

Then i is increased to M from 1, when occurring that H (i) is greater than threshold value η ₂time, current i is the coboundary of cone; Then i is increased, when occurring that H (i) is less than threshold value η ₂time, current i is the lower limb of cone;

By following formula, vertical projection is done to cone image:

$V\left(j\right)=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}F(i,j)$

Then j is increased to N from 1, when occurring that V (j) is greater than threshold value η ₂time, current j is the left hand edge of cone; Then j is increased, when occurring that V (j) is less than threshold value η ₂time, current j is the right hand edge of cone;

By horizontal and vertical projection, the edge, upper and lower, left and right of cone is demarcated, thus obtain the central coordinate of circle of cone, then according to cone size and camera resolution, select the interior external radius of annular region, annular region to be detected is split;

Step (2) carries out gradient detection to the annular region to be detected split;

Gradient detection algorithm is adopted to detect cone lumps, cone wood chip, cone white-spot defects; First adopt the gaussian filtering kernel function of 5 × 5 to carry out gaussian filtering to cone annular region to be detected, remove the impact of noise;

Grad f ' (i, j) corresponding to each point is calculated as follows to filtered annular region f:

$f^{\&prime;}(i,j)=\frac{\left|f_x^{\&prime;}(i,j)\right|+\left|f_y^{\&prime;}(i,j)\right|}{2}$

Wherein horizontal and vertical gradient is calculated as follows:

f′ _x(i,j)＝f(i+1,j)-f(i-1,j)

f′ _y(i,j)＝f(i,j+1)-f(i,j-1)

Then threshold value η is done to gradient image f ' ₃process, remove the impact of noise further:

$f^{\&prime;}(i,j)=\left\{\begin{array}{c}0,f^{\&prime;}(i,j),{\mathrm{\&eta;}}_3\\ f^{\&prime;}(i,j),\mathrm{else}\end{array}\right.$

The number of the interior non-zero points of gradient image f ' after last statistical threshold process, if not zero number is greater than threshold alpha, then judges this cone existing defects; Otherwise consistency detection is carried out to this cone region;

Step (3) is treated surveyed area and is made consistency detection;

Ring-type segmentation is carried out to the annular of cone region to be detected, only consistency detection is done to outermost two-layer annular region; Ground floor annular region is positioned at outermost, and second layer annular region is positioned at the inner side of ground floor, if each segmented areas of ground floor annular region is designated as each segmented areas of second layer annular region is designated as calculate the color average of each piecemeal in each layer annular region with color variance V _i ¹, V _i ²;

Every one deck annular region is judged as follows:

$R_i^1=\left\{\begin{array}{cc}1,& \mathrm{if}(M_i^1/(\underset{i}{\mathrm{\&Sigma;}}{M}_i^1/K)>{\mathrm{\&beta;}}_1)\mathrm{or}({V_i}^1/(\underset{i}{\mathrm{\&Sigma;}}{V_i}^1/K)>{\mathrm{\&beta;}}_2)\\ 0,& \mathrm{else}\end{array}\right.$

$R_i^2=\left\{\begin{array}{cc}1,& \mathrm{if}(M_i^2/(\underset{i}{\mathrm{\&Sigma;}}{M}_i^2/K)>{\mathrm{\&beta;}}_1)\mathrm{or}({V_i}^2/(\underset{i}{\mathrm{\&Sigma;}}{V_i}^2/K)>{\mathrm{\&beta;}}_2)\\ 0,& \mathrm{else}\end{array}\right.$

Here K represents the piecemeal number in every layer of annular region, β ₁, β ₂be respectively average and variance threshold values;

If the mark of all piecemeals is 0, then judge this cone not existing defects, otherwise judge this cone existing defects.