CN101320004A - Bamboo strip defect online detection method based on machine vision - Google Patents

Abstract

The invention discloses a bamboo splint defect online detection method based on machine vision, which comprises the steps of continuously shooting four surfaces of a bamboo splint by using a camera along the length direction of the bamboo splint to obtain a gray level image of the surface of the bamboo splint; processing the captured digital image of each frame, and judging whether defects exist on small sections of bamboo strips in the image of the frame according to a bamboo strip defect detection method; if the small sections of bamboo strips in the frame image have defects, the whole bamboo strip is considered to have defects, and meanwhile, a command of the defects is sent to a sorting mechanism; if the small sections of bamboo strips in all the frame images of the whole bamboo strip have no defects, the whole bamboo strip is considered to have no defects; while sending an instruction to the sorting mechanism that there are no defects. The invention can carry out on-line detection on the defects of the bamboo strips with different colors and shades and simultaneously detect four surfaces along the length direction of the bamboo strips.

Description

On-line detection method of bamboo strip defects based on machine vision

technical field

The invention belongs to the field of visual detection, in particular to a non-contact detection method for bamboo strip defects.

Background technique

There are 39 genera and more than 500 kinds of bamboo in my country, with a distribution area of more than 5 million hectares. Bamboo resources are abundant and have strong regeneration ability, so they are more and more widely used. After 5-10 years of general bamboo afforestation, it can be felled and utilized every year. However, due to geographical environment, climate, humidity and other reasons, some bamboos will rot from the roots during the growth process and will be infected with insects; Defects such as oblique cuts and overcuts occur, so defects such as rot, boreholes, mold, cracks and oblique cuts will appear on the bamboo strips. The production process of bamboo products is mainly composed of two steps: the first step is substrate processing, and the second step is post-processing of substrate combination. Defect detection must be carried out to the base material (ie bamboo strips) between the first and second steps, otherwise the qualified rate of bamboo products will be seriously affected. With the improvement of the automatic processing capacity of bamboo products, more and more bamboo strips need to wait for defect detection before the second step. At present, only manual inspection with naked eyes is not only greatly affected by human factors, but also various defects of different degrees It is easy to cause visual fatigue, resulting in misjudgment and missed detection of bamboo defects. At present, there is no automatic online detection method for bamboo strip defects using visual images in China.

Contents of the invention

In order to overcome the deficiency that the existing technology cannot automatically detect the defects of bamboo strips, the present invention provides a method for online detection of bamboo strip defects based on machine vision, which can detect rot, boreholes, mildew, Defects such as cracks and chamfers are automatically detected online.

The technical solution adopted by the present invention to solve its technical problems comprises the following steps:

1. Along the length direction of the bamboo strip, use the camera to continuously shoot the four surfaces of the bamboo strip to obtain the grayscale image of the bamboo strip surface; predefined: for the grayscale image, the smaller the grayscale value of the pixel, the darker the brightness, and the grayscale The larger the intensity value, the brighter the brightness;

Two. Image processing is carried out to each frame of digital images captured, and according to the bamboo strip defect detection method, judge whether there is defect on the small section bamboo strip in this frame image;

3. If there is a defect in the small section of bamboo in this frame image, then it is considered that the whole bamboo is defective, and an instruction of defect is sent to the sorting mechanism at the same time;

4. If the small bamboo strips in all frame images of the whole bamboo strip have no defects, then it is considered that this bamboo strip has no defects; at the same time, an instruction that there is no defect is sent to the sorting mechanism.

Described bamboo strip defect detection method comprises the following steps:

1. For the entire grayscale image, obtain the border of the bamboo strip through edge detection, remove the background, obtain the image f of the bamboo strip area, and copy the image data f' of the bamboo strip area, and obtain the two boundaries of the bamboo strip in the image f Straight lines L1 and L2. Calculate the angle θ between L1 and L2, if θ≥judgment threshold Tp, (Tp≤2°), it is considered that the two boundary lines of the bamboo strip are not parallel, and there is a defect of "triangular strip", go to step six, Give the detection result signal that the bamboo strips have defects; otherwise, it is considered that the two boundary lines of the bamboo strips in the image are approximately parallel, and continue to the next step of detection.

2. Perform mean filtering on the image f to remove the interference of noise points to obtain the image f1.

3. Perform a texture filtering operation on the image f1: take the current pixel as the center, and replace the gray value of the current pixel with the maximum value among the gray values of the left and right adjacent pixels in the current pixel row. The total width of the left and right neighborhood pixels is greater than or equal to half the pixel width d _H of the bamboo texture in the image, and is less than or equal to the pixel width d _B between two adjacent bamboo textures in the image (d _H and d _B are determined through experiments in advance) . All the pixels in the image f1 are processed point by point, and finally the image f2 after filtering out the bamboo texture can be obtained.

4. Detection of large defects in bamboo strips, including the following steps:

1. For the image f2, the segmentation threshold Th and the average gray value u1 and u2 of the two classes are obtained by using the maximum class distance method;

2. If the grayscale difference between u1 and u2 obtained by image f2 is less than the pre-determined experimental value D (D is pre-tested, and the qualified bamboo strip images are processed by the maximum class distance method, the obtained u1 and u2), then it is considered that the image does not contain large bamboo defects, and jumps to step five to detect small bamboo defects;

3. In the image f2, the pixel whose gray value is less than or equal to Th is the defect area. The total number of pixels num in the defect area in the image f2 and the total sum of all pixels in the image f2 are counted. If the ratio of num to sum is less than the judgment Threshold Pe (Pe is the minimum judgment threshold of the percentage of defect area in the total area of bamboo strips under the detection accuracy requirements), then the image is considered to be free of defects, and skips to step 5 to detect small defects in bamboo strips, otherwise the image is considered There is a defect in the image, and binary processing is performed on the image f2, then go to step 6, and a detection result signal of the defect in the bamboo strip is given.

The surface of the bamboo strip is lit vertically, in the bamboo strip image, the bamboo strip defects: moth hole, mildew, bamboo green area is darker than the normal area of the bamboo strip surface; parallelogram defect area and chamfering defect The area, which appears in the image as the shadow area on the side of the bamboo strip, is darker than the normal area on the bamboo strip surface. Since the above five kinds of defect areas of bamboo strips have small gray value and relatively large area in the grayscale image, it can be judged whether there are these five kinds of defects in bamboo strips through step 4.

5. Detection of small defects in bamboo strips, including the following steps:

1. For the image f' of the bamboo strip area, along the length direction of the bamboo strip, it is divided into a limited number of small areas by n pixels wide (n is less than or equal to 10);

2. Use the gradient operator to calculate the gradient value of each pixel;

3. Count the gray value Gi of the pixel corresponding to the maximum value of the gradient in each small area;

4. For Gi in all small areas, find the minimum value Gmin among them;

5. Calculate the average gray value u of the bamboo strip image f', if the difference between u and Gmin is less than d (d is the grayscale difference between u and Gmin obtained by calculating the qualified bamboo strip image in advance) , then it is considered that there is no defect in the image, go to step six, and give the detection result signal that the image has no defect;

6. Take the segmentation threshold as Gmin, and perform binary processing on the image f' to obtain a binary image f". In the image f', pixels whose gray value is less than or equal to Gmin are defective areas. For the binary image f" Labeling is performed on the defect area in , and the area and aspect ratio Scale of each defect area are calculated. If Area≥Ta and {|Scale-1|≤ξ or Scale≥Tb}, (where Ta, Tb, and ξ are respectively The total number of pixels of the smallest moth-eaten hole, the aspect ratio of the shortest crack, and the circularity range of the moth-eaten hole under the detection accuracy requirements), then it is considered that the image has defects, go to step 6, and give the defects of the bamboo strips The detection result signal; otherwise, it is considered that there is no defect in the image, and go to step six, and the detection result signal that the image has no defect is given.

For smaller moth-eaten holes, cracks and cracks, due to their small pixel width and small area, they cannot be detected and identified through step four; but the gradient values of their boundaries are large, and can be detected and identified through step five.

6. Give the detection result signal of the bamboo stick image.

The beneficial effects of the invention are: the invention can detect the defects of bamboo strips with different color depths on-line, and simultaneously detect four surfaces along the length direction of the bamboo strips. The types of bamboo strip defects that can be detected are shown in Figure 3, including: moth-eaten holes, mold, cracks, cracks, parallelograms, triangle strips, bamboo green, and oblique cuts, effectively solving the problem of bamboo strips caused by human factors. Misjudgment and missed detection of defects. The detection results are shown in Figure 4. Bamboo strip defects: moth-eaten holes, mold, cracks, cracks, parallelograms, bamboo green, and oblique cuts are shown in black in the figure, and triangular strip defects are displayed as two boundary lines of bamboo strips.

If an industrial camera with a resolution of 752×480, a frame rate of 60 frames per second, a 16mm lens, and a field of view of 70mm×45mm are used; the computer uses: CPU Celeron 2.0GHz, memory 512MB, graphics card 64M, the experimental test results are as follows Shown:

Types of Bamboo Strips Detected operation hours identification recognition error Bamboo strips with large defects 51.8～62ms 96% 4% Bamboo strips with small defects 75～82ms 92% 8% Bamboo strips without defects 80～83ms 97% 3%

The average processing time of each image is 80ms, the detection accuracy rate is 92%, and the detection speed of bamboo strips is greater than 0.87m/s.

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

Description of drawings

Fig. 1 is overall detection step flow process of the present invention;

Fig. 2 is the flow process of bamboo strip defect detection method;

Fig. 3 is 8 kinds of bamboo strip defects that the present invention can detect. In the picture: (a)-wormhole, (b)-mold, (c)-crack, (d)-crack, (e)-parallelogram, (f)-triangular strip, (g)-bamboo green, (h) - oblique cut;

Fig. 4 is the display result after the present invention detects 8 kinds of bamboo strip defects. In the picture: (a)-wormhole, (b)-mold, (c)-crack, (d)-crack, (e)-parallelogram, (f)-triangular strip, (g)-bamboo green, (h) - Miter cut.

Detailed ways

The bamboo strips detected by the invention have a width of 14mm-25mm, a thickness of 6mm-10mm, and a length of 0.9m-2.6m, and the bamboo strips are narrow and long workpieces. The light source is used to irradiate the four surfaces of the bamboo strips along the length direction vertically. In the obtained bamboo strip images, the defects of the triangular strips are shown as the two boundary straight lines of the bamboo strips have a large angle; moth holes, mildew, Compared with the normal area on the surface of the bamboo strips, the bamboo green area has darker brightness, smaller gray value, and larger area; the parallelogram defect area and the beveled defect area appear as the shadow area on the side of the bamboo strip in the image, which is similar to the The normal areas on the surface of bamboo strips are darker in brightness, smaller in gray value, and larger in area; for smaller moth-eaten holes, cracks, and cracks, they are elongated features, and the defect area is smaller, but the gradient of the boundary The value is larger.

The specific implementation is as shown in Figure 1: along the length direction of the bamboo strips, use the camera to continuously shoot the four surfaces of the bamboo strips to obtain grayscale images of the bamboo strip surfaces; image processing is performed on each frame of digital images captured, according to the bamboo strips Strip defect detection method, to judge whether there are defects on the small bamboo strips in this frame image; if there are defects in the small bamboo strips in this frame image, it is considered that the whole bamboo strip is defective, and at the same time send a defect instruction to the sorting mechanism ; If the small bamboo strips in all the frame images of the whole bamboo strip have no defects, then this bamboo strip is considered to have no defects; at the same time, an instruction of no defects is sent to the sorting mechanism.

Described bamboo strip defect detection method, as shown in Figure 2, comprises the following steps:

1. For the entire grayscale image, obtain the boundary of the bamboo strips through edge detection, remove the background, obtain the image f of the bamboo strip area, and copy the image data f' of the bamboo strip area, and obtain two lines on the surface of the bamboo strips in the image f Boundary straight lines L1 and L2. Calculate the angle θ between L1 and L2, if θ≥Tp, (Tp≤2°), it is considered that the two boundary lines of the bamboo strip are not parallel, and there is a defect of "triangular strip", go to step 6, and give the bamboo strip Otherwise, it is considered that the two borders of the bamboo strips in this image are approximately parallel, and the next step of detection is continued.

2. Perform mean filtering on the image f to remove the interference of noise points to obtain the image f1.

3. For the image f1, filter the texture: take the current pixel as the center, and replace the gray value of the current pixel with the maximum value among the gray values of the left and right neighboring pixels of the current pixel row. The total width of the left and right neighborhood pixels is greater than or equal to half the pixel width d _H of the bamboo texture in the image, and is less than or equal to the pixel width d _B between two adjacent bamboo textures in the image (d _H and d _B are determined through experiments in advance) . All the pixels in the image f1 are processed point by point, and finally the image f2 after filtering out the bamboo texture can be obtained.

4. Detection of large defects in bamboo strips, including the following steps:

1. For the image f2, the segmentation threshold Th and the average gray value u1 and u2 of the two classes are obtained by using the maximum class distance method;

2. If the gray value difference between u1 and u2 obtained from a certain image is less than the pre-determined experimental value D (D is the pre-test, the qualified bamboo strips are processed by the method of maximum class spacing, and u1 and u2 are obtained. u2), then it is considered that the image does not contain large defects of bamboo strips, and jumps to step five to detect small defects of bamboo strips;

3. In the image f2, the pixel whose gray value is less than or equal to Th is the defect area. The total number of pixels num in the bamboo defect area and the sum of all the pixels in the bamboo area in the image f2 are counted. If the ratio of hum to sum If it is less than the judgment threshold Pe (Pe is the minimum judgment threshold of the percentage of the defect area in the total area of the bamboo strips under the detection accuracy requirements), then the image is considered to be free of defects, and skips to step 5 to detect small defects in the bamboo strips, otherwise it is considered There is a defect in the image, and the image f2 is binarized, then go to step 6, and the detection result signal of the defect in the bamboo strip is given;

5. Detection of small defects in bamboo strips, including the following steps:

1. For the image f' of the bamboo strip area, the image of the bamboo strip area is divided into a limited number of small areas on average along the length direction of the bamboo strip by n pixels wide (n is less than or equal to 10);

2. Use the gradient operator to calculate the gradient value of each pixel;

3. Count the gray value Gi of the pixel corresponding to the maximum value of the gradient in each small area;

4. For Gi in all small areas, find the minimum value Gmin among them;

5. Calculate the average gray value u of the image f' of the bamboo strip area, if the difference d between u and Gmin (d is the grayscale difference between u and Gmin obtained by calculating qualified bamboo strips in advance), then it is considered There is no defect in the image, go to step 6, and give the detection result signal that the image has no defect;

6. Take the segmentation threshold as Gmin, and carry out binarization processing on the bamboo strip area image f' to obtain the image f". In the bamboo strip area image f', the pixel whose gray value is less than or equal to Gmin is the defect area. For the image The defect area in f" is labeled, and the area Area and aspect ratio Scale of each defect area are calculated. If Area≥Ta and {|Scale-1|≤ξ or Scale≥Tb}, (where Ta, Tb, ξ are respectively the total number of pixels of the smallest moth-eaten hole, the aspect ratio of the shortest crack, and the circularity range of the moth-eaten hole under the detection accuracy requirements), then it is considered that the image has defects, go to step 6, and give the existence of bamboo strips The detection result signal of the defect; otherwise, it is considered that there is no defect in the image, go to step six, and the detection result signal of the image without defect is given.

6. Give the detection result signal of the bamboo stick image.

For the bamboo strip image, after the detection step 1, it can be judged whether the bamboo strip has the defect of "triangular strip"; after the detection step 4, it can be judged whether the bamboo strip has the defect types: moth holes, mildew, bamboo green, parallelogram , Oblique cutting; after the inspection step five, the types of defects that can be judged whether the bamboo strips are: small moth-eaten holes, cracks, and cracks.

Claims (2)

1, based on the bamboo strip defect online test method of machine vision, it is characterized in that comprising the steps:

(a) along bamboo bar length direction, use camera that four surfaces of bamboo bar are taken continuously, obtain the gray level image on bamboo bar surface;

(b) each frame of digital image of catching is carried out Flame Image Process,, judge on the segment bamboo bar in this two field picture whether have defective according to the bamboo strip defect detection method;

(c) if there is defective in the segment bamboo bar in this two field picture, think that then there is defective in whole bamboo bar, send the instruction that has defective to mechanism for sorting simultaneously;

(d), think that then this root bamboo bar does not have defective if the segment bamboo bar in all two field pictures of whole bamboo bar does not all have defective; Send the instruction that does not have defective to mechanism for sorting simultaneously.

2,, it is characterized in that described bamboo strip defect detection method may further comprise the steps according to the bamboo strip defect online test method based on machine vision of claim 1:

(a) to the view picture gray level image, background is removed on the border that obtains the bamboo bar by rim detection, obtains the image f in bamboo bar zone, and duplicates the view data f ' in bamboo bar zone, obtains two the boundary straight line L1 and the L2 of bamboo bar among the image f; Calculate the angle theta between L1 and the L2, if θ 〉=decision threshold Tp, Tp≤2 ° think that then two boundary straight line of this bamboo bar are not parallel, have defective, go to step (f), provide the testing result signal that there is defective in the bamboo bar; Otherwise think that two boundary straight line of bamboo bar are approximate parallel in this width of cloth image, proceed next step detection;

(b) to image f, carry out mean filter, remove the interference of noise spot, obtain image f1;

(c) to image f1, carry out the filtering texture operation: be the center with the current pixel point,, replace the gray-scale value of current pixel point with the maximal value in the capable left and right sides neighborhood pixels gray-scale value of current pixel.The overall width of left and right sides neighborhood territory pixel is more than or equal to half pixel wide d of bamboo texture in the image _H, smaller or equal to the pixel wide d between adjacent two textures of bamboo in the image _BAll pixel pointwises among the image f1 are handled, finally can be obtained the image f2 behind the filtering bamboo texture;

(d) detection of the big defective of bamboo bar may further comprise the steps:

A. to image f2, adopt maximum kind spacing method to try to achieve average gray value u1, the u2 of segmentation threshold Th and two classes;

If B. the u1 that tried to achieve of image f2 and the gray scale difference value between the u2 think then that less than the experiment value D that measures in advance this width of cloth image does not contain the big defective of bamboo bar, jump to step (e), carry out the detection of the little defective of bamboo bar;

C. in image f2, gray-scale value is defect area smaller or equal to the pixel of Th, the total sum of all pixels among the total num of defect area pixel and the image f2 among the statistical picture f2 is if the ratio of num and sum, is then thought this width of cloth image zero defect less than decision threshold Pe, jump to step (e), carry out the detection of the little defective of bamboo bar, otherwise think that there is defective in this width of cloth image, and image f2 is carried out binary conversion treatment, go to step (f), provide the testing result signal that there is defective in the bamboo bar;

(e) detection of the little defective of bamboo bar may further comprise the steps:

A. for the image f ' in bamboo bar zone, along the length direction of bamboo bar, be divided into limited zonule by n pixel is wide, n is smaller or equal to 10;

B. use gradient operator, calculate the Grad of each pixel;

C. add up the gray-scale value Gi of gradient maximal value institute corresponding pixel points in each zonule respectively;

D. for the Gi in all zonules, find out minimum value Gmin wherein;

E. calculate the average gray u of bamboo bar area image f ',, go to step (f), provide the testing result signal that this image does not have defective if the difference of u and Gmin, thinks then that there is not defective in this width of cloth image less than the gray scale difference value d of qualified bamboo bar image;

F. getting segmentation threshold is Gmin, and image f ' is carried out binary conversion treatment, obtains bianry image f ".In image f ', gray-scale value is defect area smaller or equal to the pixel of Gmin; To bianry image f " in the defect area processing of labelling; calculate area A rea, the length breadth ratio Scale of each defect area; if Area 〉=Ta and | Scale-1|≤ξ or Scale 〉=Tb}; wherein Ta, Tb, ξ are respectively the pixel sum of the minimum borehole of accuracy of detection under requiring, the length breadth ratio, the circularity scope of borehole in short crack; think that then there is defective in this image; go to step (f), provide the testing result signal that there is defective in the bamboo bar; Otherwise think that there is not defective in this width of cloth image, go to step (f), provide the testing result signal that this width of cloth image does not have defective; (f) provide the testing result signal of bamboo bar image.

Images