KR20130087984A - Micro-crack detecting method based on improved anisotropic diffusion model using extended kernel - Google Patents

Abstract

PURPOSE: A micro-crack inspecting method based on an anisotropic diffusion model using an extensible kernel is provided to excellently detect micro-cracks even though the inspection is repeated little. CONSTITUTION: A micro-crack inspecting method based on an anisotropic diffusion model using an extensible kernel includes the following steps of: obtaining source images of a target object via machine vision (S1); calculating a basic inclination of a preset basic kernel (S3); calculating an expanded inclination of an expanded kernel of which size is expanded in comparison with the basic kernel (S4); comparing the basic inclination and the expanded inclination and calculating a diffusion coefficient according to the result of the comparison (S5-S8); obtaining diffusion images (S9); and confirming micro-cracks after subtracting the diffusion images from the source images (S10-S11). [Reference numerals] (AA) Start; (BB) Finish; (S1) Acquire source images; (S10) Source image-diffusion image; (S11) Confirm micro-cracks; (S2) Remove finger patterns; (S3) Calculate a basic inclination of a basic kernel; (S4) Calculate an expanded inclination of an expanded kernel; (S5) Basic inclination >= expanded inclination?; (S6) Inclination=basic inclination; (S7) Inclination=expanded inclination; (S8) Calculate a diffusion coefficient; (S9) Acquire the diffusion images

Description

Micro-crack Detecting method based on improved anisotropic diffusion model using extended kernel}

The present invention relates to a method for inspecting microcracks based on an anisotropic diffusion model, and more particularly, to a method for inspecting microcracks based on an anisotropic diffusion model that can accurately detect diagonal microcracks.

In general, automatic inspection of defects is performed by using a machine version in industrial sites. The automatic defect inspection is used in various manufacturing industries such as semiconductor, display, paper, and wood. (See Non-Patent Documents [1] to [6].)

Recently, many researches have been conducted on solar cells for photovoltaic power generation, and now they have reached a commercialization stage. The defect inspection of the product is performed during the manufacturing process of the solar cell, one of the important inspection items is the micro crack inspection. When there is a micro crack in the solar cell, the solar cell is broken in a manufacturing facility. In this case, it causes the interruption of the automated production line consisting of in-line, and once the production line is stopped, it takes a long time to resume. This leads to lower productivity and higher manufacturing costs. Polycrystalline members, such as the solar cell, because the pattern of the surface is very irregular, there are many difficulties in applying the existing failure detection algorithm. Therefore, using an infrared camera, a micro crack detection technique using an infrared image has been utilized. However, the price of the infrared camera is expensive, the infrared image acquisition speed is slow. Recently, in order to overcome the drawbacks described above, studies have been attempted to detect defects by applying an industrial camera having a relatively low price and a fast image acquisition speed.

FIG. 1 shows an image of an area without micro cracks and an area with micro cracks in a polycrystalline solar cell. 2 shows micro cracks in a horizontal direction and micro cracks occurring in a diagonal direction.

As described above, the micro cracks are detected by using the micro crack detection algorithm on the image obtained through the camera.

Tsai et al. Proposed a failure detection algorithm for polycrystalline solar cells using an anisotropic diffusion model (see Non-Patent Document [7]). The proposed method uses an anisotropic diffusion model. In the bad image, the microcracks are flattened, and the portions without the microcracks are subjected to a diffusion process for preserving the image, and the microcracks are detected by subtracting the bad image and the diffused bad image before diffusion. However, the research of Tsai et al. Applies only the kernels in the east, west, north, and north directions, so there is a problem in that it cannot reflect the micro crack information generated in the diagonal direction.

Ko and Rheem improved the performance of the defect detection algorithm for polycrystalline solar cells of [7] by using a dual structure anisotropic diffusion filter. The proposed method is to apply crosswise kernel and diagonal kernel in parallel.

However, as shown in FIG. 2B, when the micro crack image in the diagonal direction is discontinuously represented in the image due to the characteristics of the camera lens and the sensor, the conventional algorithm detects the micro crack in the diagonal direction discontinuously. There is a problem.

If the above two methods do not have sufficient repetitive diffusion, micro cracks are partially detected and noise pixels remain, thereby increasing the probability of misclassification in the final defect detection. However, when increasing the number of iterations for accurate detection, there is a problem that the overall algorithm execution time is increased and productivity is reduced.

[1] R.W. Conners, C.W. McMillin, K. Lin and R.E. Vasquez-Espinosa, Identifying and locating surface defects in wood, IEEE Transactions on Pattern Analysis and Machine Intelligence PAMI-5, pp. 573-583, 1983. [2] L.H. Siew and R.M. Hogdson, "Texture measures for carpet wear assessment, IEEE Transactions on Pattern Analysis and Machine Intelligence 10, pp. 92-105, 1988. [3] K.V. Ramana and B. Ramamoorthy, "Statistical methods to compare the texture features of machined surfaces," Pattern Recognition 29, pp. 1447-1459, 1996. [4] J. Escofet, M.S. Millan, H. Abril and E. Torrecilla, "Inspection of fabric resistance to abrasion by Fourier analysis," Proceedings of SPIE 3490, pp. 207-210, 1998. [5] T. Ohshige, H. Tanaka, Y. Miyazaki, T. Kanda, H. Ichimura, N. Kosaka and T. Tomoda, "Detect inspection system for patterned wafers based on the spatial-frequency filtering," IEEE / CHMT European International Electronic Manufacturing Technology Symposium, pp. 192-196, 1991. [6] K. Wiltschi, A. Pinz and T. Lindeberg, "Automatic assessment scheme for steel quality inspection," Machine Vision and Applications 12, pp. 113-128, 2000. P. Perona and J. Malik, "Scale-space and edge detection using anisotropic diffusion," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 12, no. 7, pp. 629-639, 1990. [7] D. M. Tsai, C. C. Chang and S. M. Chao, "Micro-crack inspection in heterogeneously textured solar wafers using anisotropic diffusion," Image and Vision Computing, vol. 28, no. 3, pp. 491-401, 2010. [8] J. Ko and J. Rheem, "Anisotropic Diffusion based Micro-crack Inspection in Polycrystalline Solar Wafers," Proceedings of the International MultiConference of Engineers and Computer Scientists, vol. 1, pp. 524-528, 2011. [9] P. Perona and J. Malik, "Scale-space and edge detection using anisotropic diffusion," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 12, no. 7, pp. 629-639, 1990. [10] S. Aja, C. Alberola and A. Ruiz, "Fuzzy anisotropic diffusion for speckle filtering," IEEE International Conference on Acoustics, Speech, and Signal Processing, vol. 2, pp. 1261-1264, 2001. [11] L. Zadeh and J. Kacprzyk, "Fuzzy logic for the management of uncertainty," John Wiley & Sons, Inc. 1992. [12] H. J. Zimmermann, "Fuzzy set theory and its applications," Kluwer-Nijhoff Publishing, 2001.

An object of the present invention is to provide a method for inspecting microcracks based on an anisotropic diffusion model that can more accurately detect microcracks generated in a diagonal direction.

An anisotropic diffusion model-based microcracks inspection method according to the present invention may include obtaining an original image of an inspection target through machine vision, calculating a base slope in a preset base kernel, and comparing the base kernel. Calculating an extension slope in the enlarged extended kernel, comparing the base slope with the extension slope, calculating a diffusion coefficient and constructing a diffusion image accordingly, and subtracting the diffusion image from the original image And then deriving an image for identifying the microcracks.

In addition, an anisotropic diffusion model based microcracks inspection method according to another aspect of the present invention, the step of acquiring the original image for the inspection object through machine vision, removing the finger pattern from the original image, Comprising the step of calculating the basic slope to the size 3 by (3) by the base kernel in the image, the finger pattern has been removed, and the extension slope in the 5 by 5 sized extended kernel larger than the base kernel Calculating, comparing the basic slope with the extended slope, calculating a diffusion coefficient using a relatively large value, constructing a diffusion image, subtracting the diffusion image from the original image, and then checking the microcracks. And deriving an image.

The microcracks inspection method based on the anisotropic diffusion model according to the present invention can continuously detect the microcracks in the diagonal direction.

In addition, it is possible to detect microcracks even with a small number of repetitions compared to the conventional algorithm.

In addition, it can be detected with a thickness similar to the thickness of the actual microcracks compared to the conventional algorithm.

1 is a diagram illustrating an image of a polycrystalline solar cell.
2 is a view illustrating an image in which a micro crack is enlarged.
FIG. 3 is an image comparing T and K values of micro cracks in a diagonal direction and micro cracks in a horizontal direction.
4 shows the boundary effect of the number of repetitions in the finger pattern.
5 is an image comparing before and after removing a finger pattern.
6 is a diagram illustrating a basic kernel and an extended kernel according to the present embodiment.
7 is a flowchart illustrating a method of inspecting microcracks according to the present embodiment.
8 is a resultant image according to the change of the control parameter C _pre .
9 is an image according to the result of the change of the repetition number T value and the normalized value according to the method of Tsai et al.
10 is an image according to the result of the change of the repetition number T value and the normalized value according to the method of Ko and Rheem.
11 is an image according to the result of the change in the repetition number T value and the normalization value according to the method according to the present embodiment.
12 is an image comparing the control parameter C _post by changing the micro crack.
FIG. 13 is an image comparing the results of each algorithm when T = 7, K = 30, Cpre = 1.5, and Cpost = 5.
FIG. 14 is an image comparing the results of each algorithm when T = 5, K = 30, Cpre = 1.5, and Cpost = 5.

Hereinafter, an embodiment of the present invention will be described with reference to the accompanying drawings, for example, a method for inspecting micro cracks in a polycrystalline solar cell. However, the present invention is not limited thereto.

1 is an image of a polycrystalline solar cell. FIG. 1A is an image of a multi-grain wafer surface without micro cracks, and FIG. 1B is a view of a wafer surface having micro cracks in a diagonal direction. It is a video.

Referring to FIG. 1, it can be seen that the surface is represented by an irregular texture due to the characteristics of the polycrystalline solar cell. Microcracks in polycrystalline solar cells are known to have low brightness and high gradient values in the image (see Non-Patent Document [7]). You can check it.

FIG. 2 is a view showing an enlarged image of micro cracks, and shows discontinuities of micro cracks in digital images in front-light LED illumination. FIG. 2A is an image in which micro cracks in a horizontal direction are continuous, and FIG. 2B is an image in which micro cracks in a diagonal direction are discontinuous.

Referring to FIG. 2A, micro cracks occurring in the horizontal direction are continuously represented in an image. Referring to FIG. 2B, it can be seen that the micro cracks generated in the diagonal direction are discontinuously represented without connecting the micro cracks in the diagonal direction due to the characteristics of the lens and the CCD sensor. As shown in FIG. 2B, when microcracks are discontinuously represented in an image, there is a disadvantage in that microcracks are discontinuously detected by a conventional microcracks detection algorithm.

FIG. 3 is an image comparing T and K values of micro cracks in a diagonal direction and micro cracks in a horizontal direction.

In FIG. 3, K is a parameter used in Equations 4 and 5 described later. T is the number of times to repeat the anisotropic diffusion algorithm. When subtracting the diffuse image from the original image, the diffuse image means that the diffusion process has been performed T times from the original image. Referring to FIG. 3, it can be seen that as the number of diffusions T decreases, micro cracks are discontinuously detected. In the case of detecting micro cracks discontinuously in this manner, the probability of misclassification in the final failure determination increases. Therefore, it is important to detect while maintaining the form of micro cracks.

Perona-Malik's diffusion model of Non-Patent Document [9] is a traditional anisotropic diffusion method, and is effective for smoothing the inside of a region while removing contours and removing noise. This method is a spreading filter using pixel information of a four-way cross kernel structure, and equations 1 to 4 described later are used.

Here, t denotes repetition of diffusion, and I _t (x, y) denotes a brightness value in (x, y) coordinates of the filtered image after t-stage diffusion is applied. ∇I ⁱ _t (x, y), i = 1,2,3 and 4 mean the inclination value in the east, west, north and north directions, and is expressed as in Equation 2.

In the anisotropic diffusion method, as shown in Equation 2, an image spreading process is performed by adding or subtracting neighboring pixels and a gradient value using a cross kernel. However, when the gradient value is used as it is, edge preservation is not possible because the edge information of the image is spread together. Therefore, when applying the cruciform kernel, the homogeneous region needs c ^t _i (x, y) which has an edge stopping function to reduce or stop diffusion.

In the diffusion model method of Perona-Malik, the diffusion coefficient function g (gI ⁱ _t (x, y)) is expressed as Equation 4. Hereinafter, for convenience of expression, iI ⁱ _t (x, y) will be expressed as ∇I.

Here, K is a constant, and operates as an edge strength threshold, and is a value obtained experimentally.

On the other hand, the diffusion model of Tsai et al. In the non-patent document [7] proposes a method of detecting micro cracks using information that micro cracks have low pixel brightness and high gradient values in polycrystalline solar cells. First, the part of the microcracks in the image is flattened, and the brightness value is preserved for the non-defective areas. Subsequently, when the original image and the diffused image are subtracted, the pixel value of the micro crack area is highlighted. It is a method of detecting a defect using this. In order to planarize the microcracks in the image, a diffusion coefficient function g (∇I, f) such as Equation 5 will be proposed.

Here, f (x, y) is represented by Equation 6 to be described later, and f (x, y) is a normalized value of an 8-bit gray image, and is used as a normalization parameter.

When the image is spread according to the diffusion coefficient function as described above, the spread image from the original image is subtracted as shown in Equation 7 below. Through this process, the brightness of the micro crack area is emphasized in the subtracted image.

Equation 7 shows an equation for subtracting the spread image from the original image.

Tsai et al.'S method is a model in which a diffusion equation (see Equation 5) is applied using crosswise four-direction pixel information (see Equation 2). However, the method of Tsai et al. Reflects pixel information in the east, west, north and north directions, but does not reflect the pixel information in the diagonal direction. In the polycrystalline solar cell, since the microcracks can be generated in all directions, the Tsai et al. Method has a limitation in detecting the microcracks.

Ko and Rheem of Non-Patent Document [8] are described in Tsai et al. To compensate for the shortcomings of the method, a diffusion model with a diagonal lattice structure was further applied. Equation 8 shows an expression of subtracting an original image and an image to which a conventional diffusion model of east, west, north, and north directions is applied. Equation 9 shows an equation for subtracting an original image and an image to which a diagonal diffusion model is applied.

The slope of the diagonal direction used in Equation 9 is expressed by Equation 10.

Ko and Rheem add an image to which Equation 8 and Equation 9 are applied to obtain a final image as shown in Equation 11.

As described above, the study of Tsai et al. Applied the kernels in the east, west, north and north directions, and did not reflect the micro crack information in the diagonal direction. In order to improve this problem, Ko and Rheem proposed a diffusion model of a double lattice structure to reflect diagonal micro crack information. However, there was still a disadvantage in that the micro cracks in the diagonal direction were partially detected. As shown in FIG. 2B, since the brightness of the micro cracks in the diagonal direction is not uniform due to the characteristics of the lens and the sensor, the diagonal micro cracks are partially detected by the conventional algorithm.

In order to improve the above disadvantages, in the present embodiment, the base slope of the base kernel of the original image and the extension slope of the extended kernel whose size is larger than the base kernel are respectively calculated, and the base slope and the extended slope are respectively calculated. Compare and apply the larger value to the equation for diffusion coefficient (g).

4 is a diagram illustrating a basic kernel and an extended kernel according to the present embodiment. 5 is a flowchart illustrating a method of inspecting micro cracks according to the present embodiment.

First, an original image of an inspection target is obtained through machine vision. (S1)

Pre-processing for finger pattern removal is performed on the original image. (S2)

In order to apply the research results of the non-patent documents [7] and [8] to actual machine vision applications, the finger pattern of the solar cell should be considered. Finger patterns are brightly represented in the image. When diffusion is performed including a finger pattern, a boundary effect occurs at a finger pattern boundary.

4 shows an image spread including a finger pattern region. 4A illustrates an enlarged portion of a micro crack in an image, in which a bright portion is a finger pattern, and a dark and diagonal portion is a micro crack. 4B illustrates the algorithm of Tsai et al., And as the repeated diffusion occurs, the thickness of the boundary portion of the finger pattern becomes thicker. When detecting microcracks, false detection may occur due to such a border effect. Therefore, when performing image diffusion, it should be performed except for finger patterns.

When the anisotropic diffusion model algorithm is applied to the inspection software of the actual process equipment, a teaching process is needed to distinguish the finger pattern region and the diffusion region. However, this process requires the teaching process to be repeated every time a new model solar cell is produced in a mass production environment, which reduces the overall operating time of the inspection equipment. This process also reduces the convenience of the equipment operator. Therefore, it is necessary to diffuse the entire inspection image region without distinguishing the diffusion region in the teaching process in order to improve the operation convenience and overall productivity of the operator in the solar cell inspection system. To this end, the following preprocessing is performed to replace the area of the finger pattern with the overall average brightness in the image.

First, the average brightness μ _ΔI and the standard deviation brightness σ _ΔI of the original image are calculated according to Equations 12 and 13 described later.

Thereafter, the finger pattern portion is replaced with an average value of the entire image according to Equation 14 to be described later.

After this process, if image diffusion is performed, a result similar to that without a finger pattern is obtained. In Equation 14, C _pre is a parameter for controlling how much brightness is determined as a finger pattern in a preprocessing process. Brightness of the coordinates in the image _{(ΔI 0 (x, y)} ) that is greater than the sum of the standard deviation of brightness (σ _ΔI) and C _pre multiplied by the average brightness (μ _ΔI) of, the formula for determining the finger pattern, C _pre When the brightness is greater than the average brightness (μ _ΔI ) serves to control whether to determine the finger pattern.

As described above, the present embodiment has been described as an example of performing preprocessing to remove the finger pattern after acquiring the original image. However, the present invention is not limited thereto, and the following steps may be performed without removing the finger pattern.

After removing the finger pattern as described above, the basic kernel is set to 3 by size, and the basic slope is calculated from the basic kernel. (S3) Referring to FIG. 6, a basic kernel of 3 by 3 size is set. In addition, the slope of the east, west, north and north directions is calculated according to Equation 2.

In addition, an extension kernel whose size is larger than the base kernel is set, and the extension slope is calculated by the extension kernel. Referring to FIG. 6, the extended kernel is set to 5 by 5 size, for example, but the size of the kernel is not limited thereto. The equation for obtaining the extension slope in the extension kernel is as shown in Equation 15 below.

In Fig. 6, the basic slope obtained from equation (2) and the expanded slope obtained from equation (15) are shown. The basic slope according to Equation 2 indicates an inclination of the north-west and north-west directions with the pixels of the cross-shaped structure in the center pixel (x, y), and the expanded slope according to Equation 15 has a? (#) Structure around the center pixel. It indicates the inclination of the pixels up, down, left and right. That is, the extension slope indicates a direction extending in the horizontal and vertical directions than the north, south, east and west directions with respect to the center pixel. In the case of a weak connection strength in the diagonal microcracks, the inclination of the east, west, north and south directions is insufficient. Therefore, the extended slope may be used to more actively reflect the information of the surrounding micro cracks.

Comparing the basic slope and the extended slope calculated as described above and calculating a diffusion coefficient accordingly (S5 to S7), comparing the basic slope and the extended slope, the slope having the larger value is the diffusion coefficient. Apply to obtain. That is, a larger value of the basic slope and the extended slope is applied to Equation 5. A diffusion image is derived according to the diffusion coefficient obtained from Equation 5 above (S9).

When the spread image is obtained, the spread image is subtracted from the original image.

After subtracting the diffusion image to which the diffusion model is applied from the original image, post-processing for defect detection is performed for micro crack detection. A threshold technique was applied to detect micro cracks (see Non Patent Literature [7]). This uses information of average and standard deviation of image brightness. The related formula is the same as the following formula (16). The diffused image is similar to the original image in the part without the micro crack, and the portion where the micro crack is located is flattened. When the changed diffusion image is subtracted from the original image, the portion where the micro crack is located becomes brighter, and the portion without the micro crack becomes dark. The mean and standard deviation of image brightness are used to segment only the bright areas of the subtracted image. That is, it is easy to check the microcracks by allowing only the image of the pixel whose image slope falls outside the setting range in consideration of the average brightness and the standard deviation. (S11)

Where μ _ΔI is the average brightness, σ _ΔI is the standard deviation brightness, and C _post is the adjustment parameter that adjusts the segmentation. _μΔI and _σΔI are obtained in Equations 12 and 13, respectively.

In order to confirm the performance of the algorithm according to the present embodiment, the algorithm of the existing Tsai et al. (See Non Patent Literature [7]) and Ko and Rheem's dual lattice structure algorithm (see Non Patent Literature [8]), And comparative experiments were performed with the algorithm of this example. The image acquisition conditions used in the experiments were diffused white LED surface illumination with a side angle of incidence of 45 degrees, and the image was obtained by setting the pixel resolution to 23 μm. A total of 117 images were used for the experiment. Of these, there are 96 microcracks and 21 images without defects. Each parameter used in the algorithm was selected through a repeated experiment to show the optimal performance.

5 shows the result of a pretreatment process for removing a finger pattern. FIG. 5A shows the shape of a finger pattern in an image as an original image. 5B shows the result of applying the finger pattern removal technique proposed in this embodiment. C _pre was set to 1.5 in the pretreatment process for finger pattern removal. In post-processing, C _post was set to 10. 5C shows the result of performing anisotropic diffusion without applying preprocessing and subtracting the image spread from the original image. 5D shows the result of applying preprocessing, performing anisotropic diffusion, and subtracting the diffuse image from the original image. At this time, the number of repetitions of anisotropic diffusion was set to five. In addition, K value was set to 30. In Figure 5c it can be seen that the borderline effect of the finger pattern appears. Compared to FIG. 5C, a large number of finger patterns are removed in FIG. 5D, and micro cracks are more emphasized.

8 shows the result of the change in the value of C _pre . In order to compare the effects of the change in the C _pre value, the number of diffusion repetitions T was set to 5 and K was set to 30. In addition, C _post for micro crack segmentation was set to 5.

9 is an image according to the result of the change of the repetition number T value and the normalized value according to the method of Tsai et al. 10 is an image according to the result of the change of the repetition number T value and the normalized value according to the method of Ko and Rheem. 11 is an image according to the result of the change in the repetition number T value and the normalization value according to the method according to the present embodiment.

9 to 11 illustrate the effects of the K value and the T value through the micro crack detection result according to the change of the K value and the repetition number T value. The following experiment was conducted to confirm the results of the change in K value and the number of repetitions T. K values were in the range of 10 to 50, and T was in the range of 1 to 7. C _pre was set to 1.5 and C _post was set to 5 for this experiment. 9 to 11, it can be seen that in all three methods, the noise pixels are suppressed as K increases in the same number of diffusion, and only micro cracks are detected. However, if the K value is set too large, it can be seen that the micro crack is also partially detected.

12 is an image comparing the control parameter C _post by changing the micro crack.

Referring to FIG. 12, the effect of the segmentation control constant C _post may be seen. In order to compare the effects of the change in the C _post value, the number of diffusion repetitions T was set to 5 and K was set to 30. In addition, C _pre for removing the finger pattern was set to 2.

FIG. 13 is an image comparing the results of each algorithm when T = 7, K = 30, Cpre = 1.5, and Cpost = 5. FIG. 14 is an image comparing the results of each algorithm when T = 5, K = 30, Cpre = 1.5, and Cpost = 5.

Referring to FIGS. 13 and 14, the results of performing each algorithm on the micro cracks in the diagonal direction are shown. Compared to the existing method, it can be seen that the method according to the present embodiment detects the thickness similar to that of the original microcracks. In addition, it is possible for the method according to the present embodiment to continuously detect micro cracks as compared to the existing method.

As described above, in this embodiment, an anisotropic diffusion model using an extended slope for detecting microcracks in a polycrystalline solar cell having a non-uniform background image is proposed. In order to improve the shortcomings of the microcracks detection algorithm based on the isotropic diffusion model that discontinuously detects the microcracks generated in the diagonal direction, the use of the extended slope is proposed.

A relatively large value is applied to the diffusion coefficient function by comparing the basic slope and the extended slope in the east-west-south direction. Therefore, it was confirmed that even in a portion where the connection strength of the microcracks is weakly expressed in the image, the microcracks can be effectively detected without using the surrounding pixel information. It was confirmed that the thickness of the detected microcracks does not differ from the thickness of the original microcracks even when the number of repetitions increases. In addition, it can be seen that the method of the present embodiment shows excellent performance with fewer repetitions than the conventional method. In order to confirm the performance of the algorithm of this embodiment, the experiments of the conventional Tsai et al. Algorithm, Ko and Rheem's dual lattice structure, and the algorithm of this embodiment are performed. It confirmed that it was excellent.

However, in the present embodiment, the detection of the micro cracks of the solar cell has been described by way of example.

While the present invention has been described with reference to exemplary embodiments, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

Claims (9)

Acquiring an original image of an inspection object through machine vision;
Calculating a base slope in the preset base kernel;
Calculating an extension slope in an extension kernel having a size larger than that of the base kernel;
Comparing the basic slope with the extended slope, calculating a diffusion coefficient and constructing a diffusion image accordingly;
And deriving an image for identifying a micro crack after subtracting the diffuse image from the original image. The method according to claim 1,
The default kernel is set to 3 by 3 size,
The extended kernel is an anisotropic diffusion model based microcracks inspection method is set to 5 by 5 size. The method according to claim 1 or 2,
The basic slope is an anisotropic diffusion model-based microcracks inspection method of the cross-section of the pixels of the cross-shaped structure in the basic kernel. The method according to claim 3,
The extended slope is an anisotropic diffusion model-based microcracks inspection method of the up-down, left-right slope of the pixels of the? (-) Structure in the extended kernel. The method of claim 4,
The building of the diffused image may include:
An anisotropic diffusion model based microcracks inspection method for comparing the basic slope and the extended slope to calculate a diffusion coefficient using a relatively large value. The method of claim 4,
After subtracting the diffuse image from the original image, the microcracks based on the anisotropic diffusion model that detects the microcracks by extracting only the image of the pixel whose slope is out of the setting range considering the average brightness and the standard deviation Way. Acquiring an original image of an inspection object through machine vision;
Removing a finger pattern from the original image;
Calculating a basic slope of a 3 by 3 size of a basic kernel in the image from which the finger pattern has been removed;
Calculating an extension slope in an extended kernel of 5 by 5 size extended in size than the base kernel;
Comparing the basic slope with the extended slope, calculating a diffusion coefficient using a relatively large value, and constructing a diffusion image;
And deriving an image for identifying a micro crack after subtracting the diffuse image from the original image. The method of claim 7,
After subtracting the diffuse image from the original image, the microcracks based on the anisotropic diffusion model that detects the microcracks by extracting only the image of the pixel whose slope is out of the setting range considering the average brightness and the standard deviation Way. The method of claim 1 or claim 7,
The inspection target is an anisotropic diffusion model based microcracks inspection method.

Images