CN108896574A - A kind of bottled liquor method for detecting impurities and system based on machine vision - Google Patents

Abstract

The invention discloses a method and system for detecting impurities in bottled liquor based on machine vision. Firstly, the wine liquid is moved in a flipping manner, and continuous images of the wine liquid movement are collected, and then the moving target is separated through an improved secondary difference algorithm. Finally, according to the contour properties of the moving target, it is judged whether the bottled liquor contains impurities. This detection method can not only avoid the generation of a large number of bubbles, reduce the interference caused by the bubbles, but also solve the interference caused by the background and the bottle body; therefore High-precision detection of impurities in bottled liquor can be realized.

Description

A method and system for detecting impurities in bottled liquor based on machine vision

technical field

The invention belongs to the technical field of liquor detection, and relates to a method and system for detecting impurities in bottled liquor based on machine vision.

Background technique

During the production of bottled wine, due to the production process, packaging technology and other reasons, some impurities may appear in the wine liquid. At present, the commonly used liquid impurity detection methods include artificial light detection method, semi-automatic light detection method, laser ray detection method, photoelectric (resistance) detection method, machine vision detection method, etc.; machine vision detection method, in view of its simple operation and detection accuracy With high advantages, it is widely used in beverage wine and pharmaceutical industries such as beer detection, oral liquid detection, etc.

The machine vision detection method is a non-contact detection method that combines visual sensing, digital image processing, pattern recognition and artificial intelligence technology. The moving image is processed [including filtering the image by filtering method (such as median or mean value filter) and extracting the required part by edge detection method]. Finally, the measured image processed by the computer vision system is processed, so as to obtain the Required test information (e.g. Pass or Fail). At present, the inspection equipment manufacturers based on machine vision inspection mainly include Germany Seidenada, Miho, Heuft, Italy Brevetti, GF, CMP, Japan Esia, Toshiba, etc. Specifically, impurity detection equipment based on machine vision can be divided into intermittent visual monitoring, tracking visual monitoring and impurity detection with fixed cameras. Intermittent visual inspection refers to the intermittent operation of the equipment, that is, the main wheel of the inspection adopts the "go-stop-go-stop" cycle operation mode, and the inspected product enters the camera point after passing through the high-speed rotating station, and the camera pauses for n milliseconds to obtain Continue to run after the sequence of images. This type of representative equipment includes Italian Brevetti small-capacity drug inspection machine, Italian CMP light inspection machine, etc. The main inspection volume is between 1-20mL, and the speed is generally 150-300 bottles/min. Similar models have the advantages of stable image acquisition and relatively simple mechanical devices, but at the same time, the detection speed is slow, and it is easy to cause misjudgment or missed detection for large-capacity product detection. Tracking visual inspection equipment means that the inspection wheel runs at a constant speed, and the imaging device tracks and takes pictures of the inspected products, acquires a sequence of images, and then returns quickly to track and take pictures of the next batch of inspected products. The main feature of this type of detection system is to detect the continuous operation of the main wheel. Through the operation of the camera at the same angular speed or the operation of the mirror, the product image is relatively static in the sequence image, but the impurities have relative displacement, thus judging the positive and negative Taste. This type of equipment mainly includes Filtec, Miho, Seidenada, Brevett, Pharmamech, Bosch, etc. Most manufacturers at home and abroad use tracking visual inspection. Due to the use of tracking shooting, the detection speed is faster than intermittent, generally at 200-600 bottles/minute , The detection capacity is in the range of 1 to 500 ml. The advantage of the tracking method is that it is fast, and the bottle burst rate is low, but its mechanical device is more complicated and the tracking camera needs to be accurate and stable. The impurity detection of the fixed camera method uses a fixed camera. When the tested product passes by, take a small amount of photos and compare them with the standard photos stored in the program, so as to judge whether there are obvious foreign objects in the tested product. This method is the fastest and can fully meet the requirements. On-line inspection requirements on high-speed automated production lines, but due to the algorithmic idea of comparing with standard photos, the detection accuracy is not high, and only obvious foreign objects in the solution can be identified.

In addition, for transparent liquid impurities such as oral liquids and injections, the impurity detection method based on machine vision mainly adopts the method of rotating emergency stop, using the camera to obtain continuous frame images, then processing the images to find moving targets, and finally using support vector machines Or other tracking algorithms (such as meanshift algorithm) to identify foreign matter and complete the detection of impurities. However, this impurity detection method based on machine vision is mainly aimed at the detection of colored liquids such as beer, health wine, oral liquid, etc., and has not yet specifically targeted at the detection of impurities in bottled liquor; especially based on the rotating emergency stop method, it is easy to A large number of bubbles will be generated during the process, which seriously affects the detection accuracy of impurities in liquor.

Therefore, the existing liquid impurity detection method based on machine vision is not perfect enough to meet the needs of the liquor industry for impurity detection.

Contents of the invention

The object of the present invention is to provide a method for detecting impurities in bottled liquor based on machine vision, so as to realize accurate detection of impurities in bottled liquor.

Another object of the present invention is to provide a system for detecting impurities in bottled liquor based on machine vision.

Aiming at bottled liquor, especially round transparent bottled liquor, the present invention provides a method for detecting impurities in bottled liquor based on machine vision. First, continuous multi-frame images are obtained, and then multi-frame images are processed to obtain images containing moving objects. Then complete the detection of impurities in bottled liquor according to the contour shape of the moving target. The present invention further uses an improved quadratic difference algorithm to realize the problem of segmenting small targets from a large number of interference backgrounds by utilizing the continuity of sequence images in time and space.

The bottled liquor impurity detection method based on machine vision provided by the invention comprises the following steps:

(1) Turn over the bottled liquor 180°;

(2) video collection of bottled liquor after flipping;

(3) Extract N frames of images from the continuous frame images of the video, M frames of images are separated between each frame of images and the previous frame of images or the next frame of images, wherein N is a positive odd number, and M is an integer greater than or equal to 0;

(4) Preprocessing the extracted N frames of images to remove image noise;

(5) Using the quadratic difference algorithm to analyze the preprocessed image, and extract the image containing the moving target;

(6) Extract the contour of the moving target, and calculate the ratio of the contour major axis to the minor axis;

(7) Comparing the obtained contour long-short axis ratio with the set threshold, if the contour long-short axis ratio is 1.0-1.1, the moving target is a bubble; if the contour long-short axis ratio is greater than 1.4, the moving target is an impurity.

In the method for detecting impurities in bottled liquor based on machine vision, in order to reduce a large number of air bubbles caused by the movement of the liquid, the present invention adopts a method of turning over 180° to make the liquor move. Compared with the commonly used rotary emergency stop method, this method can realize the movement of the liquid in the bottle and possible impurities while reducing the generation of air bubbles. Especially when using an inverted 180° (that is, first turn the bottle body upside down, the bottle mouth is super-lower, and then turn the bottle body 180°), the impurities can be located near the bottleneck, and the inversion can make the impurities near the central axis as much as possible, thereby reducing Missing detection occurs due to impurities close to the bottle wall.

The above-mentioned detection method of impurities in bottled liquor based on machine vision has been obtained through repeated experiments. The light source and lighting method used for detection are related to the type of impurities. Videos are collected from the front of bottled liquor. When the detected impurities are white, dark The lighting method of background and bottom lighting; when the detected impurities are black, the lighting method of back lighting is used; this is conducive to obtaining high-quality video images. For round wine bottles, it is easy to produce light reflection and refraction effects, resulting in a lot of interference around the bottle wall; in order to prevent interference due to the physical characteristics of round wine bottles, set A layer of filter paper to reduce the interference of light on imaging.

In the method for detecting impurities in bottled liquor based on machine vision, in order to ensure the effect of image processing and improve the detection accuracy of impurities in bottled liquor, the video contains at least three consecutive frame images.

The above method for detecting impurities in bottled liquor based on machine vision improves detection efficiency while ensuring the detection accuracy of impurities in bottled liquor. The N is 3 or 5, and the value of M is: 2≤M≤5.

The above method for detecting impurities in bottled liquor based on machine vision, in order to eliminate background interference and remove image noise, preprocess the extracted N frames of images to remove image noise, including the following sub-steps:

(41) Carry out Gamma correction to the proposed N frame image, and convert the image after Gamma correction into a grayscale image;

(42) Process the grayscale image obtained in step (41) by using a weighted mean filtering algorithm to filter out image noise.

In the above step (41), Gamma correction is performed on the extracted N frames of images to adjust the contrast of bright and dark areas (for example, reduce the contrast of highlighted areas), so as to more effectively reduce the interference caused by light reflection and refraction effects. Conventional Gamma correction methods already disclosed in the art can be used. Then convert the Gamma-corrected image into a grayscale image.

The purpose of above-mentioned step (42) is to remove image noise, can adopt conventional filter algorithm disclosed in the art to carry out filtering process to image and remove noise, what the present invention adopts is based on the weighted mean filter algorithm of median value (Li Xiufeng, Su Lanhai, Rong Hui Fang, Chen Hua. Research on Improved Mean Filtering Algorithm and Its Application[J]. Microcomputer Information, 2008(01): 235-236+202.) Filtering the obtained grayscale image, retaining while removing image noise Detect the target, so that the detection result is more accurate. Above-mentioned bottled liquor impurity detection method based on machine vision, the present invention adopts quadratic difference algorithm to process the image after step (4) pretreatment, comprises the following sub-steps:

(51) In the N frame of images, two adjacent frames of images are differentially calculated to obtain (N-1) differential images;

(52) In the (N-1) difference images, every two adjacent difference images are divided into one group, and the two difference images in each group are respectively subjected to difference operation and energy accumulation operation to obtain the associated secondary Difference image and energy accumulation image;

(53) Subtract the obtained energy accumulation image from its associated quadratic difference image to obtain an image containing a moving target.

By adopting the quadratic difference algorithm, the moving target in the bottle can be separated, and the interference caused by the static object images such as the background and the bottle body can be solved.

In the above method for detecting impurities in bottled liquor based on machine vision, in step (5), the extracted images containing moving objects are subjected to binarization threshold processing using Otsu's maximum inter-class variance method to effectively separate the background image and eliminate the background as much as possible interference; and use the morphological opening operation 2×2 to remove the bright spots less than 4 pixels in the image.

In the above-mentioned method for detecting impurities in bottled liquor based on machine vision, the long axis and short axis of the contour of the moving target refer to the length and width of the rectangle circumscribing the contour of the moving target respectively.

The above method for detecting impurities in bottled liquor based on machine vision, in order to be able to distinguish impurities and air bubbles, classifies impurities and air bubbles according to their shape characteristics. When the ratio of the long and short axes of the contour of the moving target is 1.0 to 1.1, the moving target is determined to be a bubble; when the moving target When the ratio of the major axis to the minor axis of the contour is greater than 1.1 and less than or equal to 1.4, it may be a bubble or an impurity. Therefore, in order to improve the accuracy of impurity detection, the moving target with a contour major axis ratio greater than 1.4 is determined as an impurity.

When the distinction between impurities and air bubbles cannot be realized based on the contour of the moving target, the above steps (1) to (7) can be repeated for re-detection.

The present invention further provides a machine vision-based impurity detection system for bottled liquor, comprising:

The image acquisition device is used to collect video of bottled liquor turned over 180°;

The image processing device is used to process the video collected by the image acquisition device to complete the detection of impurities in bottled liquor; the image processing device includes:

The frame image extraction unit extracts N frames of images from the continuous frame images of the video, and there are M frames of images between each frame image and the previous frame image or the next frame image, wherein N is a positive odd number, and M is greater than or equal to 0 positive integer;

An image preprocessing unit is used to preprocess the extracted N frames of images to remove image noise;

The moving target image extraction unit processes the pre-processed image by using the quadratic difference algorithm to extract the image containing the moving target;

The contour extraction and long-short axis ratio calculation unit extracts the contour of the moving target and calculates the ratio of the long axis to the short axis of the contour;

The judging unit compares the obtained contour long-short axis ratio with the set threshold, if the contour long-short axis ratio is 1.0-1.1, the moving target is a bubble; if the contour long-short axis ratio is greater than 1.4, the moving target is an impurity.

In the above-mentioned bottled liquor impurity detection system based on machine vision, the detection system is provided with a black impurity detection station and/or a white impurity detection station, and an image acquisition device is configured at the black impurity detection station and/or white impurity detection station; For the black detection station, the image acquisition window of the image acquisition device is aimed at the bottled liquor turned over 180°, and at the same time, the light source is used to give light from the back of the bottled liquor; Bottled liquor, at the same time, use the light source to give light from the bottom of the bottled liquor, and use a dark background on the back of the bottled liquor. Since the black impurities are easier to detect and the detection accuracy is higher, the black impurities are detected first, and then the white impurities are detected, which can further improve the detection accuracy while improving the detection efficiency. Therefore, when setting the black impurity detection station and the white detection station, first put the bottled liquor into the black detection station for detection, if it contains black impurities, the bottled liquor will be screened out directly, without entering the next white impurity detection station If there is no black impurity, enter the white impurity detection station to judge whether it contains white impurities.

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

(1) The present invention first adopts the overturning method to make the wine liquid move, and collects the continuous images of the wine liquid motion, then separates the moving target through the improved secondary difference algorithm, and finally judges whether the bottled liquor contains Impurities, this detection method can not only avoid a large number of bubbles and reduce the interference caused by the bubbles, but also solve the interference caused by the background and the bottle; therefore, it can achieve high-precision detection of impurities in bottled liquor;

(2) The present invention puts the bottled liquor in an upside-down state before turning over, so that the impurities are located near the bottleneck, and after turning over, the impurities are as close to the central axis as possible, further reducing the situation of missed detection due to the impurities being close to the bottle wall;

(3) The present invention can respectively realize the detection of black and white impurities in bottled liquor, and improve the detection efficiency while ensuring the detection accuracy of impurities;

(4) The present invention is carrying out data processing to the continuous image of bottled white wine liquor collected, and in the process of extracting the moving target, by setting the filter paper behind the light source and the Gamma correction in the later stage, the light reflection and light caused by the light source can be effectively reduced refraction interference;

(5) The present invention further adopts Otsu's maximum inter-class variance method to carry out threshold value processing to the extracted image containing the moving target, which can effectively separate the background and further reduce the interference of the background;

(6) The present invention can be realized by relying on existing equipment, is simple and efficient in operation, and is suitable for popularization and use in the field of detection of liquid impurities, especially bottled liquor impurities.

Description of drawings

Fig. 1 is a schematic flow chart of machine vision-based detection of white impurities in bottled liquor according to Embodiment 1 of the present invention.

Fig. 2 is a schematic diagram of detection of impurities in bottled liquor based on machine vision in the present invention; wherein (a) a schematic diagram of bottled liquor upside down, (b) a schematic diagram of detection of white impurities in bottled liquor, (c) a schematic diagram of detection of black impurities in bottled liquor; among the figures, 1- Video camera, 2-bottled white wine, 3-red filter paper, 4-LED light source, 5-black shading plate.

FIG. 3 is a schematic flow chart of extracting a moving target using a quadratic difference algorithm in Embodiment 1 of the present invention.

Fig. 4 is a schematic diagram of the detection effect of bottled liquor impurities based on machine vision in the present invention; wherein (a), (e), (i) and (i1) respectively correspond to the collection image, secondary difference processing image, and motion of fiber impurities in bottled liquor Enlarged image of target outline and moving target outline, (b), (f), (j) and (j1) correspond to the collected image, secondary difference processing image, moving target outline and moving target outline enlargement of black slag impurities in bottled liquor, respectively Figures, (c), (g), (k) and (k1) respectively correspond to the collected image of glass debris in bottled liquor, the second difference processing image, the outline of the moving target and the enlarged image of the moving target outline, (d), ( h), (l) and (l1) respectively correspond to the collected image of the white floc impurity in bottled liquor, the image processed by the second difference, the contour of the moving target and the enlarged image of the contour of the moving target.

Fig. 5 is a schematic flow chart of the detection of black impurities in bottled liquor based on machine vision according to Embodiment 2 of the present invention.

FIG. 6 is a schematic flow chart of extracting a moving target using a quadratic difference algorithm in Embodiment 2 of the present invention.

Detailed ways

The embodiments of the present invention will be given below in conjunction with the accompanying drawings, and the technical solutions of the present invention will be further clearly and completely described through the embodiments. Apparently, the embodiments described are only some of the embodiments of the present invention, but not all of them. Based on the contents of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative work belong to the protection scope of the present invention

Example 1

In this embodiment, the detection process of impurities in bottled liquor based on machine vision is described in detail by taking the detection of white impurities such as glass shavings and fiber wool as an example.

The bottled liquor white impurity detection process provided by the present embodiment, as shown in Figure 1, comprises the following steps:

(1) Turn over the bottled liquor 180°.

Due to various types of interference such as scales, patterns, adsorbed impurities on the body of bottled liquor, and fine lines caused by collisions during production and transportation, it is difficult to detect tiny foreign objects in liquor. As shown in Figure 2(a), first turn the bottled liquor 2 containing white impurities upside down, and turn the bottled liquor 2 180° when entering the area to be detected. In this inverted way of entering, the interference of the wine bottle is in multiple images The middle is relatively static, but the liquid in the bottle and possible impurities move from near the neck of the bottle along the central axis of the bottled liquor 2 due to the upside-down of the bottle body, and a sequence of images is obtained. This can not only reduce the occurrence of missed detection due to impurities close to the bottle wall, but also generate fewer air bubbles compared with the existing rotary emergency stop method, thereby minimizing the interference of air bubbles on impurity detection and improving the accuracy of impurity detection.

(2) Collect video of the flipped bottled liquor.

In this embodiment, a high-definition camera 1 is used to shoot a video of the overturned bottled liquor 2 . In order to improve the detection accuracy of white impurities, as shown in Figure 2 (b), when detecting white impurities in this embodiment, the back of the bottled liquor 2 adopts a black light-shielding plate 5 to form a dark background, and the LED light source 4 is illuminated from the bottom of the bottled liquor 2. At the same time, a layer of red filter paper 3 is added in front of the LED light source 4 (the used high-definition digital camera has a better effect on red light), thereby reducing the interference of light on imaging.

(3) Extract 3 frames of images from the continuous frame images of the video, and each frame of images is separated from the previous frame of images or the next frame of images by 2 frames of images, so that while ensuring that the moving target is extracted from the images, Reduce errors caused by uncertain factors caused by interference. Those skilled in the art can properly adjust the number of frames between two adjacent frame images according to the moving speed of the impurity.

(4) Preprocessing the extracted 3 frames of images to remove image noise, including the following sub-steps:

(41) Carry out conventional Gamma correction to the 3 frames of images mentioned, reduce the contrast of the highlighted area, and convert the image after Gamma correction into a grayscale image;

(42) Process the grayscale image obtained in step (41) by adopting a median-based weighted mean filtering algorithm to filter out image noise while retaining the target to be measured.

In this embodiment, the weighted average filtering algorithm based on the median comprises the following steps:

(A) The input image is represented by X=[x(i,j)] _M×N , where x(i,j) represents the gray value of the pixel at point (i, j) in the image gray value matrix, W ₃ (i,j) represents a window with a center pixel at (i,j) and a size of 3×3 (3×3 is chosen because the impurity is small, and an appropriate window can be selected according to the estimated impurity size in the field). but

(B) First use W ₃ (i,j) to scan in X=[x(i,j)] _M×N , and read the gray value x(i,j) of each pixel in W ₃ (i,j) ), and sort the gray values by size, and take the maximum value MAX and the minimum value MIN in the gray value.

(C) If x(i, j) is equal to the maximum value MAX or the minimum value MIN, then set the multiplication weight to 0, that is, regardless of their influence on the image, and then according to the formula

A＝x ₁ (i,j)×m ₁ +x ₂ (i,j)×m ₂ +x ₃ (i,j)×m ₃ +x ₄ (i,j)×m ₄ +x ₅ (i ,j)×m ₃ +x ₆ (i,j)×m ₂ +x ₇ (i,j)×m ₁

Obtain A, where x ₁ (i,j), x ₂ (i,j)...x ₇ (i,j) are the gray values arranged from small to large except the maximum value MAX or minimum value MIN, m ₁ , m ₂ , m ₃ , and m ₄ are the weights corresponding to each gray value, and the weight value selection rules are as follows: m ₄ corresponds to the median value of these gray values, and the coefficient is the largest, m ₃ , m ₂ , m ₁ gradually decreases, and m ₁ , m ₂ , m ₃ , and m ₄ all belong to (0,1), m ₁ , m ₂ , m ₃ , and m ₄ can be set according to the above rules.

(D) order

And assign the Z value to the pixel center point x(i, j)=Z in the area scanned by the window, that is.

Image noise can be filtered out by processing the obtained grayscale image according to the above steps (A)-(D).

(5) Using the quadratic difference algorithm to analyze the preprocessed image, and extract the image containing the moving target.

In this step, an improved quadratic difference algorithm is used to analyze the preprocessed image to separate tiny moving objects, as shown in Figure 3, which specifically includes the following steps:

(51) Perform difference operation on two adjacent frames of images among the three frames of images to obtain two difference images.

Use f(x,y,t-2), f(x,y,t), f(x,y,t+2) to represent the three frames of images processed in step (4), the first frame image f(x , y, t-2) and the second frame image f(x, y, t) are separated by 2 frames of images, the second frame image f(x, y, t) and the third frame image f(x, y, There are 2 frames of images between t+2). The absolute difference between the first frame image f(x,y,t-2) and the second frame image f(x,y,t), the second frame image f(x,y,t) and the third frame image f( x, y, t+2) do the absolute difference, and get two difference images d _{(t-2, t)} (x, y) and d _{(t, t+2)} (x, y); here you can directly Call the function absdiff() in OpenCV to complete the operation.

d _(t-2,t) (x,y)=|f(x,y,t)-f(x,y,t-2)|

d _(t,t+2) (x,y)=|f(x,y,t)-f(x,y,t+2)|

(52) Perform differential operation on the two difference images again to obtain a second difference image, and perform energy accumulation operation on the two difference images to obtain an energy accumulation image.

In this embodiment, two difference images d _{(t-2, t)} (x, y) and d _{(t, t+2)} (x, y) are made absolute difference to obtain image D (x, y); Directly call the function absdiff() in OpenCV to complete the operation.

D(x,y)=|d _(t-2,t) (x,y)-d _(t,t+2) (x,y)|.

In this embodiment, two differential images d _{(t-2, t)} (x, y) and d _{(t, t+2)} (x, y) are energy accumulated to obtain an image E (x, y); Directly call the function addWeighted(src1,scr2,1,result) in OpenCV to complete the operation, where src1 and src2 are images d _(t-2,t) (x,y) and d _(t,t+2) ( x,y).

E(x,y)=d _(t-2,t) (x,y)+d _(t,t+2) (x,y).

(53) Subtract the obtained energy accumulation image from its associated quadratic difference image to obtain an image containing a moving target.

In this embodiment, the image E(x, y) is subtracted from the image D(x, y) to obtain the grayscale image F(x, y) containing the moving object. It can be done by directly calling the function addWeighted(src1,scr2,-1,result) in OpenCV, where src1 and src2 are image E(x,y) and image D(x,y) respectively.

F(x,y)=E(x,y)-D(x,y).

Therefore, only the enhanced moving target in the middle frame is retained in the obtained image, which reduces the interference of the bottle itself.

In order to eliminate background interference as much as possible, this embodiment further uses Otsu's maximum inter-class variance method to perform binarization threshold processing on the extracted images containing moving objects, so as to effectively separate the background image; Bright spots smaller than 4 pixels in an image.

(6) Extract the contour of the moving target, and calculate the ratio of the long axis to the short axis of the contour.

In this embodiment, the long axis and short axis of the moving target contour refer to the length and width of the circumscribed rectangle of the moving target contour respectively.

(7) Comparing the obtained contour long-short axis ratio with the set threshold, if the contour long-short axis ratio is 1-1.1, the moving target is a bubble; if the contour long-short axis ratio is greater than 1.4, the moving target is an impurity.

In order to be able to distinguish impurities and air bubbles, the classification is carried out according to the shape characteristics of impurities and air bubbles. When the ratio of the long axis to the short axis of the moving target contour is 1.0 to 1.1, the moving target is determined to be a bubble; when the ratio of the long axis to the short axis of the moving target contour is greater than 1.1 and When it is less than or equal to 1.4, it may be a bubble or an impurity. Therefore, in order to improve the accuracy of impurity detection, the moving target whose contour long-short axis ratio is greater than 1.4 is determined as an impurity.

When the distinction between impurities and air bubbles cannot be realized based on the contour of the moving target, the above steps (1) to (7) can be repeated for re-detection.

In order to illustrate the detection effect of the method for detecting white impurities in bottled liquor based on machine vision provided in this example, fiber wool, glass chips and white flocs are added to bottled liquor as white impurities, and the above steps (1)-( 7) Detect the white impurities, and obtain corresponding collected images (Figure 4(a), (c) and (d)), secondary difference processing images (Figure 4(e), (g) and (h)), The contours of moving targets (Fig. 4(i), (k) and (l)) and the enlarged contours of moving targets (Fig. 4(i1), (k1) and (l1)). From Figure 4(i1), the ratio of the major axis to the minor axis of the moving target profile is 6.3, so it is determined that the moving target is an impurity; from Figure 4(k1), the ratio of the major axis to the minor axis of the moving target profile is 1.6, so it is determined that the moving target is an impurity; as shown in Figure 4 (l1), the ratio of the major axis to the short axis of the moving target contour is 2.0, so it is determined that the moving target is an impurity. The above analysis results are basically consistent with the set impurity size. Therefore, accurate detection of white impurities in bottled liquor can be realized through the detection method provided in this embodiment.

Example 2

In this embodiment, the detection process of impurities in bottled liquor based on machine vision is described in detail by taking the detection of black impurities in black slag as an example.

The bottled liquor black impurity detection process that present embodiment provides, as shown in Figure 5, comprises the following steps:

(1) Turn over the bottled liquor 180°.

Due to various types of interference such as scales, patterns, adsorbed impurities on the body of bottled liquor, and fine lines caused by collisions during production and transportation, it is difficult to detect tiny foreign objects in liquor. As shown in Figure 2(a), first turn the bottled liquor 2 containing white impurities upside down, and turn the bottled liquor 2 180° when entering the area to be detected. In this inverted way of entering, the interference of the wine bottle is in multiple images The middle is relatively static, but the liquid in the bottle and possible impurities move from near the neck of the bottle along the central axis of the bottled liquor 2 due to the upside-down of the bottle body, and a sequence of images is obtained. This can not only reduce the occurrence of missed detection due to impurities close to the bottle wall, but also generate fewer air bubbles compared with the existing rotary emergency stop method, thereby minimizing the interference of air bubbles on impurity detection and improving the accuracy of impurity detection.

(2) Collect video of the flipped bottled liquor.

In this embodiment, a high-definition camera 1 is used to shoot a video of the overturned bottled liquor 2 . In order to improve the detection accuracy of black impurities, as shown in Figure 2(c), in this embodiment, when detecting black impurities, the LED light source 4 is illuminated from the back of the bottled liquor 2, and the backlighting method is adopted.

(3) Extract 5 frames of images from the continuous frame images of the video, and there are 5 frames of images between each frame of images and the previous frame of images or the next frame of images, so that while ensuring that the moving target is extracted from the images, Reduce errors caused by uncertain factors caused by interference. Those skilled in the art can properly adjust the number of frames between two adjacent frame images according to the moving speed of the impurity.

(4) Preprocessing is carried out to the extracted 5 frames of images to remove image noise, including the following sub-steps:

(41) Carry out conventional Gamma correction to the 5 frames of images mentioned, reduce the contrast of the highlighted area, and convert the image after Gamma correction into a grayscale image;

(42) Process the grayscale image obtained in step (41) by adopting a median-based weighted mean filtering algorithm to filter out image noise while retaining the target to be measured.

In this embodiment, the weighted average filtering algorithm based on the median comprises the following steps:

(A) The input image is represented by X=[x(i,j)] _M×N , where x(i,j) represents the gray value of the pixel at point (i, j) in the image gray value matrix, W ₃ (i,j) represents a window with a center pixel at (i,j) and a size of 3×3 (3×3 is chosen because the impurity is small, and an appropriate window can be selected according to the estimated impurity size in the field). but

(B) First use W ₃ (i,j) to scan in X=[x(i,j)] _M×N , and read the gray value x(i,j) of each pixel in W ₃ (i,j) ), and sort the gray values by size, and take the maximum value MAX and the minimum value MIN in the gray value.

(C) If x(i, j) is equal to the maximum value MAX or the minimum value MIN, then set the multiplication weight to 0, that is, regardless of their influence on the image, and then according to the formula

A＝x ₁ (i,j)×m ₁ +x ₂ (i,j)×m ₂ +x ₃ (i,j)×m ₃ +x ₄ (i,j)×m ₄ +x ₅ (i ,j)×m ₃ +x ₆ (i,j)×m ₂ +x ₇ (i,j)×m ₁

Obtain A, where x ₁ (i,j), x ₂ (i,j)...x ₇ (i,j) are the gray values arranged from small to large except the maximum value MAX or minimum value MIN, m ₁ , m ₂ , m ₃ , and m ₄ are the weights corresponding to each gray value, and the weight value selection rules are as follows: m ₄ corresponds to the median value of these gray values, and the coefficient is the largest, m ₃ , m ₂ , m ₁ gradually decreases, and m ₁ , m ₂ , m ₃ , and m ₄ all belong to (0,1), m ₁ , m ₂ , m ₃ , and m ₄ can be set according to the above rules.

(D) order

And assign the Z value to the pixel center point x(i, j)=Z in the area scanned by the window, that is.

Image noise can be filtered out by processing the obtained grayscale image according to the above steps (A)-(D).

(5) Using the quadratic difference algorithm to analyze the preprocessed image, and extract the image containing the moving target.

In this step, an improved quadratic difference algorithm is used to analyze the preprocessed image and separate tiny moving objects, as shown in Figure 6, which specifically includes the following steps:

(51) Perform difference operation on two adjacent frames of images among the five frames of images to obtain two difference images.

Take f(x,y,t-10),f(x,y,t-5),f(x,y,t),f(x,y,t+5),f(x,y,t +10) represents the five frames of images processed in step (4), and there is an interval of 5 frames of images between two adjacent frames of images. The absolute difference is made between two adjacent frames of images to obtain four difference images (difference image 1 to difference image 4); here, the operation can be completed by directly calling the function absdiff() in OpenCV.

(52) The four difference images are divided into two groups, and the two difference images in each group are respectively subjected to difference operation and energy accumulation operation to obtain associated secondary difference images and energy accumulation images.

In this embodiment, the absolute difference between the difference image 1 and the difference image 2 is obtained to obtain the second difference 1 of the image, which can be completed by directly calling the function absdiff() in OpenCV. Energy accumulation of differential image 1 and differential image 2 to obtain energy accumulation 1 can be completed by directly calling the function addWeighted(src1,scr2,1,result) in OpenCV, where src1 and src2 are differential image 1 and differential image respectively 2

In this embodiment, the absolute difference between the difference image 3 and the difference image 4 is obtained to obtain the image secondary difference 2, which can be completed by directly calling the function absdiff() in OpenCV. Energy accumulation of differential image 3 and differential image 4 to obtain energy accumulation 2 can be completed by directly calling the function addWeighted(src1, scr2, 1, result) in OpenCV, where src1 and src2 are differential image 3 and differential image respectively 4.

(53) Subtract the obtained energy accumulation image from its associated quadratic difference image to obtain an image containing a moving target.

In this embodiment, the energy accumulation 1 is subtracted from the secondary difference 1 to obtain the image 1 containing the moving object, and the energy accumulation 2 is subtracted from the secondary difference 2 to obtain the image 2 containing the moving object. These can be obtained by directly calling OpenCV The function addWeighted(src1,scr2,-1,result) in the operation is completed, where src1 is energy accumulation 1 or energy accumulation 2, and scr2 is secondary difference 1 or secondary difference 2.

Therefore, in the obtained image 1 containing the moving object and the image 2 containing the moving object, only the enhanced moving object in the middle frame is retained, which reduces the interference of the bottle body itself. The image 2 containing the moving target and the image 1 containing the moving target are the movement states of the same moving target at different times. Detection can increase the accuracy of judgment.

In order to eliminate background interference as much as possible, this embodiment further uses Otsu's maximum inter-class variance method to perform binarization threshold processing on the extracted image 1 containing the moving object and image 2 containing the moving object, so as to effectively separate the background image; and use Morphological opening operation 2×2 removes bright spots smaller than 4 pixels in the image.

(6) Extract the contour of the moving target, and calculate the ratio of the long axis to the short axis of the contour.

In this embodiment, any one of the image 1 containing the moving object and the image 2 containing the moving object is the research object, and the long axis and short axis of the contour of the moving object are calculated.

(7) Comparing the obtained contour long-short axis ratio with the set threshold, if the contour long-short axis ratio is 1-1.1, the moving target is a bubble; if the contour long-short axis ratio is greater than 1.4, the moving target is an impurity.

In order to be able to distinguish impurities and air bubbles, the classification is carried out according to the shape characteristics of impurities and air bubbles. When the ratio of the long axis to the short axis of the moving target contour is 1.0 to 1.1, the moving target is determined to be a bubble; when the ratio of the long axis to the short axis of the moving target contour is greater than 1.1 and When it is less than or equal to 1.4, it may be a bubble or an impurity. Therefore, in order to improve the accuracy of impurity detection, the moving target whose contour long-short axis ratio is greater than 1.4 is determined as an impurity.

When the distinction between impurities and air bubbles cannot be realized based on the contour of the moving target, the above steps (1) to (7) can be repeated for re-detection.

In this embodiment, black slag is used as black impurities, and the obtained corresponding collected images, secondary difference processing images, moving target contours and moving target contour enlarged images are respectively shown in Figure 4 (b), (f), (j), (j1) , given in Figure 4(j1), the ratio of the major axis to the minor axis of the moving target profile is 1.7, so it is determined that the moving target is an impurity. The above analysis results are basically consistent with the set impurity size. Therefore, accurate detection of black impurities in bottled liquor can be realized through the detection method provided in this embodiment.

Example 3

This embodiment provides a system for detecting impurities in bottled liquor based on machine vision, which can detect black impurities and white impurities in bottled liquor. The detection system includes:

Black impurity detection station, including bottled liquor bottle support parts and LED light source, LED light source provides light from the back of bottled liquor;

White impurity detection station, including bottled liquor bottle support parts, LED light source, red filter paper and black shading plate, the part of the wine bottle support part aligned with the bottom of the wine bottle is a transparent structure to facilitate light transmission; the LED light source from The bottom of the bottled liquor is illuminated, and a red filter paper is set in front of the LED light source; a black light-shielding plate is placed on the back of the bottled liquor, so that the back of the bottled liquor is a dark background;

The image acquisition device is used to collect video of bottled liquor turned over 180°, which can be a high-definition digital camera, CCD industrial camera, etc.;

The image processing device is used to process the video collected by the image acquisition device to complete the detection of impurities in bottled liquor; the image processing device includes:

The frame image extraction unit extracts N frames of images from the continuous frame images of the video, and there are M frames of images between each frame image and the previous frame image or the next frame image, wherein N is a positive odd number, and M is greater than or equal to 0 positive integer;

An image preprocessing unit is used to preprocess the extracted N frames of images to remove image noise;

The moving target image extraction unit processes the pre-processed image by using the quadratic difference algorithm to extract the image containing the moving target;

The contour extraction and long-short axis ratio calculation unit extracts the contour of the moving target and calculates the ratio of the long axis to the short axis of the contour;

The judging unit compares the obtained contour long-short axis ratio with the set threshold, if the contour long-short axis ratio is 1.0-1.1, the moving target is a bubble; if the contour long-short axis ratio is greater than 1.4, the moving target is an impurity.

Both the black impurity detection station and the white impurity detection station can be equipped with an image acquisition device, or only one image acquisition device can be used. Detect switching between stations. Since the black impurities are easier to detect and the detection accuracy is higher, the black impurities are detected first, and then the white impurities are detected, which can further improve the detection accuracy while improving the detection efficiency. Therefore, when setting the black impurity detection station and the white detection station, first put the bottled liquor into the black detection station for detection, if it contains black impurities, the bottled liquor is directly screened out, and there is no need to enter the next white impurity detection station If there is no black impurity, enter the white impurity detection station to judge whether it contains white impurities.

The above-mentioned image processing device composed of frame image extraction unit, image preprocessing unit, moving target image extraction unit, contour extraction and long-short axis ratio calculation unit, and determination unit can be loaded into a computer or a processor with image processing function to complete the image processing. Processing and impurity judgment and other functions.

Claims (10)

1. a kind of bottled liquor method for detecting impurities based on machine vision, it is characterised in that include the following steps：

(1) bottled liquor is overturn 180 °；

(2) video is acquired to the bottled liquor after overturning；

(3) N frame image is extracted from the sequential frame image of video, between each frame image and previous frame image or next frame image It is spaced M frame image, wherein N is positive odd number, and M is the positive integer more than or equal to 0；

(4) the N frame image of extraction is pre-processed, removes picture noise；

(5) it is analyzed using the image that second order difference algorithm obtains pretreatment, extracts the image containing moving target；

(6) profile of moving target is extracted, and calculates the ratio of profile long axis and short axle；

(7) obtained profile ratio of long axis to short axis is compared with the threshold value of setting, if profile ratio of long axis to short axis belongs to 1.0~1.1, Moving target is bubble；If profile ratio of long axis to short axis is greater than 1.4, moving target is impurity.

2. the bottled liquor method for detecting impurities according to claim 1 based on machine vision, it is characterised in that when to black When colored foreign detects, the bottled liquor after overturning is adopted using back to light, while to the bottled liquor after overturning in step (2) Collect video.

3. the bottled liquor method for detecting impurities according to claim 1 based on machine vision, it is characterised in that work as dialogue When colored foreign detects, use dark background, bottom to light the bottled liquor after overturning in step (2), while to the bottle after overturning It fills white wine and acquires video.

4. special according to claim 1 to the bottled liquor method for detecting impurities described in 3 any claims based on machine vision Sign is the sequential frame image in the video comprising at least three frames.

5. the bottled liquor method for detecting impurities based on machine vision according to claim 4, it is characterised in that the N is 3 Or 5, the value of the M is：2≤M≤5.

6. the bottled liquor method for detecting impurities based on machine vision according to claim 1, it is characterised in that the step (4) include it is following step by step：

(41) Gamma correction is carried out to the N frame image that lifts, and the image after Gamma is corrected is converted into gray level image；

(42) it is handled using the gray level image that weighted mean filter algorithm obtains step (41), filtering image noise.

7. the bottled liquor method for detecting impurities based on machine vision according to claim 1, it is characterised in that step (5) packet Include it is following step by step：

(51) by N frame image, adjacent two field pictures carry out calculus of differences and obtain (N-1) width difference image；

(52) by (N-1) width difference image, the adjacent two width difference image of every two is divided into one group, two width difference diagrams in every group As progress calculus of differences and energy accumulation operation obtain associated second order difference image and energy accumulation image respectively；

(53) obtained energy accumulation image second order difference image subtraction associated therewith is obtained containing moving target Image.

8. according to claim 1 or the 7 bottled liquor method for detecting impurities based on machine vision, it is characterised in that step (5) in, binarization threshold processing is carried out using Otsu ' s maximum variance between clusters to the image containing moving target of extraction, and The bright spot in image less than 4 pixels is removed with morphology opening operation 2 × 2.

9. a kind of bottled liquor Impurity Detection System based on machine vision, it is characterised in that including：

Image collecting device, for acquiring video to 180 ° of overturning of bottled liquor；

Image processing apparatus is handled for the video to image acquisition device, completes the inspection to bottled liquor impurity It surveys；Described image processing unit includes：

Frame image extraction unit, extracts N frame image from the sequential frame image of video, each frame image and previous frame image or under M frame image is spaced between one frame image, wherein N is positive odd number, and M is the positive integer more than or equal to 0；

Image pre-processing unit pre-processes the N frame image of extraction, removes picture noise；

Movement destination image extraction unit is handled using the image that second order difference algorithm obtains pretreatment, extracts and contain There is the image of moving target；

Contours extract and ratio of long axis to short axis computing unit, extract the profile of moving target, and calculate the ratio of profile long axis and short axle Value；

Judging unit compares obtained profile ratio of long axis to short axis with the threshold value of setting, if profile ratio of long axis to short axis belongs to 1.0 ~1.1, moving target is bubble；If profile ratio of long axis to short axis is greater than 1.4, moving target is impurity.

10. the bottled liquor Impurity Detection System based on machine vision according to claim 9, it is characterised in that the detection system System is provided with black defects inspecting station and/or white defects inspecting station, and black defects inspecting station and/or white defects inspecting station are matched Set an image collecting device；

For black detection station, the Image Acquisition window alignment of image collecting device overturns 180 ° of bottled liquor, while benefit With light source from bottled liquor back to light；

For white detection station, the Image Acquisition window alignment of image collecting device overturns 180 ° of bottled liquor, while benefit With light source from bottled liquor bottom to light, and bottled liquor back uses dark background.