CN109087370B - A method for generating images of spongy defects in castings - Google Patents

Abstract

The invention discloses a method for generating a spongy defect image of a casting, comprising the steps of: (1) randomly generating a spongy defect skeleton image, the step is to define an initial skeleton first, then randomly generate branch skeletons according to a preset number of skeletons, and determine the skeleton Then, each skeleton randomly generates an offset compared to the reference curve, and randomly sets the initial gray level for all skeletons; the upper and lower lines in each skeleton are randomly offset by a certain distance, and the gray level between the upper and lower lines is gray. (2) filter the skeleton image; (3) superimpose noise in the filtered image; (4) superimpose the noise-added image on the casting image to obtain a spongy defect image. The defects generated by the method of the invention are diverse, the shape and size can be controlled by parameters, and have certain randomness, and the image effect is relatively close to the real image, and can be accepted as a defect image sample through the subjective judgment of experts.

Description

A method for generating images of spongy defects in castings

technical field

The invention relates to the technical field of defect image generation, in particular to a method for generating a spongy defect image of a casting.

Background technique

In the actual production of casting, although the defective rate has been greatly reduced by improving the casting process and materials, the unqualified castings cannot be completely eliminated. It is still necessary to inspect the internal defects of each casting product to ensure the delivery of the products. the quality of.

At present, automatic defect detection has gradually become the demand of the foundry industry, and a good detection algorithm often needs a large number of defect samples for training and testing. The low defective rate will make it difficult to collect some defect samples in large quantities, and can generate a large number of defects similar to real defects. The algorithm of the simulated image becomes an important solution to increase the sample size. Since there are many types of casting defects, such as shrinkage porosity, inclusions, bubbles, spongy, etc., the methods of each defect generation are different. Among them, sponge-like defects are one of the types of casting defects defined in international standards. After casting, the material is not dense enough to form a material shrinkage similar to a sponge shape, and its shape, pores and cracks are very different.

At present, there are two main types of defect image generation methods. One is based on CAD software, and the depth map of casting perspective is calculated according to the ray attenuation coefficient, that is, the projection map of the defect is calculated according to the intensity of ray absorption. Due to the great arbitrariness and randomness of the shape of the defect, this method needs to manually define the shape of the defect, and then calculate the absorption intensity of the ray according to the shape. Therefore, the shape of each defect needs to be designed manually, which will extremely The application of this method is largely limited to the large-scale generation of shapes with randomly changing properties. The other is based on the X-ray image feature analysis of the defect, using a non-parametric method different from the vector image calculation to generate a variety of typical features of the defect, and then through the feature fusion and background fusion, and finally generate a defect image on the workpiece. . However, at present, the defect images obtained by this method generally have the problems that the form is relatively simple, and the image effect is far from the real image. It cannot be regarded as a sample for the training and testing of the subsequent algorithm.

Therefore, it has important research significance and practical value to study a method that can quickly realize the generation of spongy defect images and make the generated defects more realistic.

SUMMARY OF THE INVENTION

The purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art, and to provide a method for generating images of spongy defects in castings. The method generates defects with diversity, whose shape and size can be controlled by parameters, and have certain randomness. The image effect is relatively close to the real image, and it can be accepted as a defective image sample through the subjective judgment of experts.

The object of the present invention is achieved through the following technical solutions: a method for generating a spongy defect image of a casting, comprising the steps:

(1) Randomly generate a sponge-like defect skeleton image, the steps are to define an initial skeleton first, then randomly generate branch skeletons according to the preset number of skeletons, determine the reference curve of the skeleton, and then randomly generate each skeleton compared with the reference curve Offset, the initial grayscale is randomly set for all skeletons; the upper and lower lines in each skeleton are randomly offset by a certain distance, and the grayscale between the upper and lower lines is randomly adjusted by a difference;

(2) Filter the skeleton image;

(3) Superimpose noise in the filtered image;

(4) The image after adding noise is superimposed on the casting image to obtain a spongy defect image.

Preferably, in the step (1), the initial skeleton length is defined, that is, the number of pixels represented by the skeleton in the longitudinal direction; the randomly generated branch skeleton length is less than or equal to one tenth of the initial skeleton length. Thereby, the effect of branching can be more constructed.

Preferably, in the step (1), it is assumed that each skeleton randomly generates an offset b _i (1) compared to the reference curve, and the offset is the abscissa of the starting position and the reference curve in the same row. Difference, the value range is an integer between -L ₁ /10 and L ₁ /10, L ₁ represents the initial skeleton length, that is, the starting coordinate of the i-th branch skeleton is (x _yi +b _i (1), y _i ), where x _yi is the abscissa of the reference curve on the yi _th row, and y _i represents the starting ordinate of the ith branch skeleton.

Preferably, in order to ensure the authenticity of the image, the offset of the next row in each skeleton is the offset of the previous row plus a random integer between -m1 and m2, and its grayscale is also the grayscale of the previous row Plus a random integer between -n1 and n2. Here m1, m2, n1, and n2 are empirical values, which can be obtained based on comprehensive consideration of the shape, size, grayscale, etc. of the defects in the currently collected image.

Preferably, in the step (2), Gaussian filtering is performed on the skeleton image according to the characteristics of the spongy defect image.

Preferably, in the step (3), a random Perlin noise image is generated according to the non-zero pixel positions of the filtered image, and then the noise image is superimposed on the filtered image to obtain an image after adding noise.

Further, after Gaussian filtering is performed on the skeleton image, in order to control the grayscale of the image and make the image overlay result more like an actual defect, the grayscale value of the filtered image is mapped to between 0 and G, and the random Perlin noise is The image is mapped between 0 and a, and then the two mapped images are superimposed to obtain an image after adding noise. Here G and a are empirical values, which can be set according to the characteristics of the current actual image.

Preferably, in the step (4), a superposition position is first selected on the casting image, and then the image after adding noise is rotated at a certain angle and then superimposed on the above position.

Compared with the prior art, the present invention has the following advantages and beneficial effects:

1. In the process of generating the sponge-like defect skeleton, the shape and size of the present invention can be directly controlled by parameters, and at the same time, it also has a certain randomness, which makes the generated defects diverse, and does not depend on the actual situation in terms of grayscale or other geometric shapes. Defect sample.

2. According to the characteristics of the generated spongy defect skeleton and the actual casting spongy defect image, the present invention selects a further processing method combining Gaussian filtering and Perlin noise, and the processed image is closer to the real image. According to the subjective judgment of the eye, it has been accepted as a defective image sample.

3. In the present invention, the processing method of constructing a skeleton, then filtering and adding noise, the amount of computation is small, and the speed of generating samples is fast.

Description of drawings

FIG. 1 is a flow chart of generating a spongy defect image of a casting in this embodiment.

FIG. 2 is a reference graph generated in this example.

FIG. 3 is a skeleton diagram of a randomly generated spongy defect in this embodiment.

FIG. 4 is a diagram after Gaussian filtering is performed on the skeleton in this embodiment.

FIG. 5 is a Perlin noise map randomly generated in this embodiment.

FIG. 6 is a diagram of spongy defects generated in this example.

FIG. 7 is a comparison diagram of the spongy defect map generated in this embodiment and the actual spongy defect map.

Detailed ways

The accompanying drawings are for illustrative purposes only, and should not be construed as limitations on this patent; in order to better illustrate the present embodiment, some parts of the accompanying drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art It is understandable to the artisan that certain well-known structures and descriptions thereof may be omitted from the drawings. The present invention will be described in further detail below with reference to the embodiments and the accompanying drawings, but the embodiments of the present invention are not limited thereto.

Example

As shown in Figure 1, a method for generating a spongy defect image of a casting in the present embodiment includes the following steps:

1. Randomly generate spongy defect skeleton images.

In this embodiment, according to the characteristics of the actual casting sponge-like defect image, a sponge-like defect skeleton is constructed, and the defect image is subsequently generated on the basis of the skeleton. This skeleton is the biggest innovation point of the present invention, and the concrete steps are as follows:

(1) Select an appropriate number of skeletons N. The first skeleton can be used as the initial skeleton, and the rest can be used as branch skeletons. In this embodiment, N is a random number between 40 and 80.

(2) Determine the initial skeleton length L ₁ , that is, the number of pixels in the longitudinal direction of the initial skeleton. In this embodiment, L ₁ =400.

(3) Randomly generate the starting ordinates _yi of all branch skeletons, and the lengths _Li . In this embodiment, the length of the branch skeleton is less than or equal to one tenth of the length of the initial skeleton.

(4) Draw the reference curve of the skeleton according to the actual situation of the application, which can be an arc as shown in Figure 2, or an arbitrary curve segment or a straight line segment.

(5) Randomly generate the starting offset b _i (1) of all skeletons, the offset is the difference between the starting position and the abscissa of the reference curve in the same line, and the value range is -L ₁ /10 to An integer between L _1/10 . That is, the starting coordinate of the i-th skeleton is (x _yi +b _i (1), y _i ), where is the abscissa of the x _yi reference curve on the yi- _th row.

(6) Randomly set initial grayscale a _i (1) for all skeletons, that is, the grayscale of (x _yi +b _i (1), y _i ) pixels in the image.

(7) Draw all the skeletons line by line from top to bottom. The offset of the next line of the i-th skeleton is the offset of the previous line plus a random integer between -2 and 2, and its grayscale is also the previous line. Grayscale plus a random integer between -10 and 10. The final drawn sponge-like defect skeleton image results are shown in Figure 3.

Second, filter the skeleton image.

In this embodiment, Gaussian filtering is performed on the skeleton image of the spongy defect, as shown in FIG. 4 , and its gray value is mapped between 0 and 20.

Third, superimpose noise in the filtered image.

Based on the non-zero pixel locations of the filtered image, a random Perlin noise image is generated, as shown in Figure 5. Then map the noise image between 0 and 5, and superimpose it on the filtered image to obtain the image after adding noise as shown in Figure 6.

4. Rotate the image after adding noise to an appropriate angle, and then add and superimpose it on the appropriate position of the casting image to obtain a spongy defect image.

Figure 7 includes the spongy defect map generated in this embodiment and the actual spongy defect map. After comparing the two, it can be seen that the spongy defect map generated by the method of this embodiment is relatively close to the actual spongy defect, and it can be are identified as samples for training and testing of subsequent algorithms.

The above-mentioned embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited by the above-mentioned embodiments, and any other changes, modifications, substitutions, combinations, The simplification should be equivalent replacement manners, which are all included in the protection scope of the present invention.

Claims (3)

1. A method for generating an image of a spongy defect of a casting is characterized by comprising the following steps:

(1) randomly generating a spongy defect skeleton image, wherein the steps of defining an initial skeleton, randomly generating branch skeletons according to a preset skeleton number, determining a reference curve of the skeleton, randomly generating an offset for each skeleton compared with the reference curve, and randomly setting initial gray levels for all the skeletons; the upper and lower lines of each framework are randomly offset for a certain distance, and the gray scale between the upper and lower lines is randomly adjusted by a difference value;

(2) performing Gaussian filtering on the skeleton image;

(3) generating a random Berlin noise image according to the non-zero pixel position of the filtered image, and then superposing the noise image on the filtered image to obtain an image added with noise; after the skeleton image is subjected to Gaussian filtering, mapping the gray value of the filtered image to be between 0 and G, mapping a random Berlin noise image to be between 0 and a, wherein G, a is an empirical value, and then overlapping the two mapped images to obtain an image added with noise;

(4) superposing the image added with the noise on the casting image to obtain a spongy defect image;

in the step (1), it is set that each skeleton randomly generates an offset b compared with a reference curve_i(1) The offset is the difference between the initial position and the abscissa of the reference curve on the same row, and the value range is-L₁10 to L₁Integer between/10, L₁Denotes the initial skeleton length, i.e. the starting coordinate of the ith branch skeleton is (x)_yi+b_i(1),y_i) Wherein x is_yiAs a reference curve at the y-th_iAbscissa of the row, y_iRepresenting the initial ordinate of the ith branch skeleton; the offset of the next row in each skeleton is the offset of the previous row plus a random integer between-m 1 and m2, the gray scale of the next row is also the gray scale of the previous row plus a random integer between-n 1 and n2, and m1, m2, n1 and n2 are all empirical values.

2. The casting spongy defect image generation method according to claim 1, wherein in the step (1), an initial skeleton length, namely the number of pixels in the longitudinal direction of the skeleton, is defined; the randomly generated branch skeleton length is less than or equal to one tenth of the initial skeleton length.

3. The method for generating the image of the spongy defect of the casting as recited in claim 1, wherein in the step (4), a superposition position is selected on the image of the casting, and then the image with the noise is rotated by a certain angle and is superposed on the position.

Images