CN110490192A - A kind of commodity production date tag detection method and system - Google Patents

Abstract

The invention relates to a product production date label detection method and system. The system includes a processor, a bracket, an industrial camera, a ring light source, a conveyor belt and a stage; the bracket and the conveyor belt are arranged on the stage, and the industrial camera is arranged on the conveyor belt through the bracket Above, when the commodity to be detected passes under the lens of the industrial camera via the conveyor belt, the processor controls the industrial camera to capture the image and starts the software system identification algorithm. First, the target image is thresholded using HSV space and grayscale space. The suspicious character connected domain is screened out by the scientific features; a rapid segmentation and detection scheme of the production date label with rotation is proposed to detect the production date label on the upper surface of the commodity. The invention can effectively eliminate the interference features with the same size and shape as the target character image, and can quickly segment and detect the character target.

Description

A kind of product production date label detection method and system

technical field

The invention relates to the technical field of non-contact intelligent measurement and control, and more particularly, to a method and system for detecting the production date label of commodities.

Background technique

Production date label defect detection technology belongs to computer vision technology. It is an engineering discipline that captures objects with cameras, obtains image information, and then performs image analysis to obtain information such as the pose and shape of objects. Compared with traditional manual inspection and detection, computer vision is free from artificial factors, and has the advantages of fast, accurate, non-contact and low cost, which greatly reduces labor costs and improves production efficiency.

The detection of objects in computer vision often requires image segmentation, the purpose of which is to separate the object to be detected from other signals such as background and noise. In the segmentation process, automatic threshold segmentation methods such as fixed threshold segmentation method, edge segmentation method or maximum inter-class variance can be used. No matter which segmentation method is used, it cannot guarantee that the target and the interference signal are completely separated, which requires that the detection scheme must be aimed at the situation of the interference signal.

In addition, when the production date label is printed on the surface of the product, its rotation direction is random, and at the same time, there are interference features in the image whose size and grayscale are close to the character target, which are highly recommended for the segmentation and detection of production date labels. requirements.

The invention provides a detection system and a method for a product production date label with a rotation, which can effectively eliminate the interference feature and realize the rapid segmentation and detection of the production date label of the product with a rotation.

SUMMARY OF THE INVENTION

In order to overcome the defect of the above-mentioned prior art that label detection cannot exclude interference features, the present invention provides a product production date label detection method and system.

The method includes:

S1: Obtain the RGB image of the product surface, and preprocess the RGB image to obtain a binary image; among them, RGB is the color representing the three channels of red, green, and blue, and this standard includes almost all colors that can be perceived by human eyesight. , is one of the most widely used color systems.

S2: Perform connected domain screening on the binary image obtained by S1; obtain the connected domain set of the filtered suspicious targets.

S3: Perform fast segmentation processing of the production date label with rotation on the connected domain set of the suspicious target, and obtain an image T _out of the production date character combination.

S4: establish a character comparison library; obtain the images of the characters on the production date label through experiments, and form an image set CH={CH _i }, i=1, 2, 3, ..., n; the characters Including 0~9, a~z, A~Z, ".", "/", "\" and "-".

S5: Cut out each connected domain in the image T _out , and perform knn prediction and recognition in turn; and obtain the prediction result.

Preferably, S1 includes the following steps:

S1.1: Obtain the RGB image O of the product surface and its grayscale image G; convert the image O to the HSV space (Hue, Saturation, Value) to obtain the image H; perform a fixed threshold segmentation on the image G, due to the target object and detection The environment is fixed, and its segmentation threshold can be obtained by multiple experiments to obtain a binary image B, the target pixel is set to 255, and the background pixel is set to 0;

Among them, HSV (Hue, Saturation, Value) is a color space created by A.R. Smith in 1978 according to the intuitive characteristics of color, also known as the Hexcone Model. The parameters of the color in this model are: Hue ( H), saturation (S), lightness (V).

S1.2: Let any pixel of the image H be Hi, and its HSV component is {h, _s , v _} ; let the pixel in the image B with the same pixel position as Hi be _Bi ; if _Hi and _Bi satisfy Condition Con, set B _i to 0, where Condition Con is:

Among them, R _h , R _s , and R _v represent the range of H, S, and V component values of the target pixel; because the target object and the detection environment are fixed, the interval range of R _h , R _s , and R _v is obtained by multiple experiments; traverse the images H and Image B, obtain a new obtained binary image B.

Preferably, S2 includes the following steps:

S2.1: Detect all connected domains in the binary image B after preprocessing, and obtain a connected domain set Re={Re _i },i=1,2,3,...,n;

S2.2: Set the connected domain subset of the suspicious target as Rc, and detect the smallest circumscribed rectangle of the connected domain Re _i , whose long side length is a and the short side length is b. If a∈R _a and b∈R _b , then Save Re _i in Rc, where R _a and R _b represent the interval range of the length of the long side and the length of the short side of the minimum circumscribed rectangle of the connected domain, respectively, and R _a and R _b can be obtained through multiple experiments;

S2.3: Traverse the connected domain set Re to obtain the filtered connected domain subset Rc of the suspicious target, and the filtered connected domain includes production date characters and some interference connected domains with similar shapes to the production date characters;

Preferably, S3 includes the following steps:

S3.1: Calculate the center coordinates of the smallest circumscribed rectangle of each connected domain in the connected domain subset Rc, and obtain the center coordinate set C={C _i }, i=1, 2, 3,..., n, and represent the position of the corresponding element in Rc with the coordinate set;

S3.2: Define the set N={N _i }, i=1, 2, 3, ......, n; let i=0, j=0;

S3.3: Take the element C _i as the center, select a rectangular area Rect_w whose width and height are w (the pixel length of the production date string, obtained through multiple experiments), and count the number of elements in the set C that the Rect_w area contains Number n, if n>t (where t is a preset threshold, which can be obtained by multiple experiments), then let N _j =C _i , j=j+1, otherwise start the next step;

S3.4: Execute i=i+1; and judge whether C _i ∈ C is established, if so, go back to step S3.3, otherwise start the next step;

S3.5: Establish a solution pool for selecting j elements from the set N according to the permutation and combination:

P={P _k },k=1,2,3,...,n

That is, the set P contains all the cases where j elements are selected from the set N, and each element P _k in the set P represents a situation where j elements are selected from the set N, and each element of the set P is different from each other. ; let k = 0;

S3.6: adopt the scheme P _k to select j elements from the set N, form the j elements into a whole T, and calculate the ratio f of the length of the short side to the long side of the minimum circumscribed rectangle Rect_T of T;

S3.7: Set the threshold range of the ratio interval to ratio, that is, the range of the ratio of the short side to the long side of the minimum circumscribed rectangle of the connected domain. The ratio interval is determined according to the character size. Since the detection target is fixed, its value can be obtained by multiple experiments. f∈ratio, then get the rotation angle of Rect_T Then start step S3.10, otherwise let k=k+1, and then start the next step;

S3.8: If P _k ∈ P, go back to step S3.6, otherwise set j=j-1, and then start the next step;

S3.9: If j>n _t , go back to step S3.5, where n _t is determined by the number of target characters, otherwise output segmentation failure information, and end the segmentation process;

S3.10: Cut out the overall image of the elements in the Rect_T area in the connected domain set Rc, and according to the rotation angle Rotate it until the long side of the image coincides with the horizontal direction to obtain the image T _out of the combination of production date characters.

Preferably, S5 is specifically: cutting out each connected domain in the image T _out , performing KNN (k-Nearest Neighbor, K adjacent algorithm) prediction in turn, and then comparing the prediction result with the probability threshold pn, the probability threshold pn is determined by multiple times According to the experiment, if the predicted probability is greater than pn, the predicted result is used as the character value of the connected domain, otherwise it is determined that the connected domain character is defective.

Among them, KNN prediction is to calculate the distance between the test data and the database data, sort the distance from small to large, select the k database data with the smallest distance, and use it as the prediction result;

The invention also provides a detection system applying the method for detecting the production date label of a commodity, the system comprising a stand, a conveyor belt, a bracket, an industrial camera, a light source, and a processor;

The conveyor belt is set on the stand; it is used to convey the goods to be tested;

One end of the bracket is fixed on the side of the stand, and the other end is used to fix the industrial camera, so that the industrial camera is located above the conveyor belt, and the industrial camera lens is facing down, which is used to capture the image of the production date label of the goods to be inspected conveyed on the conveyor belt, and will capture The image information is transmitted to the processor;

The light source is arranged under the industrial camera to supplement the light source for the industrial camera;

The processor controls the industrial camera to capture images, and performs data processing on the images captured by the industrial cameras, including using HSV space and grayscale space to perform threshold segmentation on the target image, and filtering out suspicious character connected domains based on grayscale features and morphological features .

Preferably, the light source is a ring light source.

Compared with the prior art, the beneficial effects of the technical solution of the present invention are: compared with the traditional manual detection, the automatic detection of the product production date label has the advantages of rapidity, accuracy, non-contact and low cost, and greatly reduces the labor cost. Increase productivity.

The rapid segmentation algorithm of the production date label with rotation proposed by the invention effectively solves the detection difficulty caused by the randomness of the rotation direction of the production date label when the production date label is printed on the surface of the product.

The rapid segmentation algorithm of the production date label with rotation proposed by the invention can effectively eliminate the interference features whose size and gray scale are close to the character target, and can quickly segment and detect the character target.

Description of drawings

FIG. 1 is a flow chart of a method for detecting a production date label of a commodity.

Figure 2 is an image of the connected domain set of suspicious objects after morphological screening.

Fig. 3 is a schematic diagram of the center coordinates of the corresponding connected domain in Fig. 2;

Fig. 4 is the flow chart of fast segmentation algorithm;

Fig. 5 is the target character set image after rotation;

FIG. 6 is a schematic structural diagram of a product production date label detection system.

In the figure, 1-processor, 2-stand, 3-industrial camera, 4-lens, 5-light source, 6-conveyor, 7-product to be inspected, 8-stand.

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 this embodiment, some parts of the drawings are omitted, enlarged or reduced, which do not represent the size of the actual product;

It will be understood by those skilled in the art that some well-known structures and their descriptions may be omitted from the drawings.

The technical solutions of the present invention will be further described below with reference to the accompanying drawings and embodiments.

Example 1:

The present embodiment provides a method for detecting the production date label of a commodity, as shown in FIG. 1 , the method includes:

S1: Obtain the RGB image of the product surface, and preprocess the RGB image to obtain a binary image;

S2: Perform connected domain screening on the binary image obtained by S1; obtain the connected domain set of the filtered suspicious targets;

S3: perform rapid segmentation processing of the production date label with rotation on the connected domain set of the suspicious target, and obtain the image T _out of the production date character combination;

S4: Establish a character comparison library; obtain character images of special characters such as 0-9, a-z, A-Z, ".", "/", "\" and "-" on the production date label through experiments, and form an image Set CH={CH _i },i=1,2,3,...,n;

S5: Cut out each connected domain in the image T _out , and perform KNN prediction and recognition in turn; and obtain the prediction result.

S1 includes the following steps:

S1.1: Obtain the RGB image O of the product surface and its grayscale image G; convert the image O to the HSV space (Hue, Saturation, Value) to obtain the image H; perform a fixed threshold segmentation on the image G (due to the target object and detection The environment is fixed, and its segmentation threshold can be obtained by multiple experiments), and a binary image B is obtained, the target pixel is set to 255, and the background pixel is set to 0;

S1.2: Let any pixel of the image H be Hi, and its HSV component is {h, _s , v _} ; let the pixel in the image B with the same pixel position as Hi be _Bi ; if _Hi and _Bi satisfy Condition Con, set B _i to 0, where Condition Con is:

Among them, R _h , R _s , and R _v represent the range of H, S, and V component values of the target pixel; because the target object and the detection environment are fixed, the interval range of R _h , R _s , and R _v is obtained by multiple experiments; traverse the images H and Image B, obtain a new obtained binary image B.

S2 includes the following steps:

S2.1: Detect all connected domains in the binary image B after preprocessing, and obtain a connected domain set Re={Re _i },i=1,2,3,...,n;

S2.2: Set the connected domain subset of the suspicious target as Rc, and detect the smallest circumscribed rectangle of the connected domain Re _i , whose long side length is a and the short side length is b. If a∈R _a and b∈R _b , then Save Re _i in the connected domain set Rc, where R _a and R _b represent the interval range of the length of the long side and the length of the short side of the minimum circumscribed rectangle of the connected domain, respectively, and R _a and R _b can be obtained through multiple experiments;

S2.3: Traverse the connected domain set Re, and obtain the connected domain subset Rc of the filtered suspicious targets. As shown in Figure 2, the filtered connected domain includes the production date characters and some interference connections with similar shapes to the production date characters. area;

As shown in Figure 4, S3 includes the following steps:

S3.1: Calculate the center coordinates of the smallest circumscribed rectangle of each connected domain in the connected domain subset Rc, and obtain the center coordinate set C={C _i }, i=1, 2, 3,..., n, display in the image as shown in Figure 3, and use this coordinate set to represent the position of the corresponding element in Rc;

S3.2: Define the set N={N _i }, i=1, 2, 3, ......, n; let i=0, j=0;

S3.3: Take the element C _i as the center, select a rectangular area Rect_w whose width and height are w (the pixel length of the production date string, obtained through multiple experiments), and count the number of elements in the set C that the Rect_w area contains Number n, if n>t (where t is a preset threshold, which can be obtained by multiple experiments), then let N _j =C _i , j=j+1, otherwise start the next step;

S3.4: Execute i=i+1; and judge whether C _i ∈ C is established, if so, go back to step S3.3, otherwise start the next step;

S3.5: Establish a solution pool for selecting j elements from the set N according to the permutation and combination:

P={P _k },k=1,2,3,...,n

That is, the set P contains all the cases where j elements are selected from the set N, and each element P _k in the set P represents a situation where j elements are selected from the set N, and each element of the set P is different from each other. ; let k = 0;

S3.6: adopt the scheme P _k to select j elements from the set N, form the j elements into a whole T, and calculate the ratio f of the length of the short side to the long side of the minimum circumscribed rectangle Rect_T of T;

S3.7: Set the threshold range of the ratio interval to ratio, that is, the range of the ratio of the short side to the long side of the minimum circumscribed rectangle of the connected domain. The ratio interval is determined according to the character size. Since the detection target is fixed, its value can be obtained by multiple experiments. f∈ratio, then get the rotation angle of Rect_T Then start step S3.10, otherwise let k=k+1, and then start the next step;

S3.8: If P _k ∈ P, go back to step S3.6, otherwise set j=j-1, and then start the next step;

S3.9: If j>n _t , go back to step S3.5, where n _t is determined by the number of target characters, otherwise output segmentation failure information, and end the segmentation process;

S3.10: Cut out the overall image of the elements in the Rect_T area in the connected domain set Rc, and according to the rotation angle Rotate it until the long side of the image coincides with the horizontal direction, as shown in Figure 5, to obtain the image T _out of the combination of production date characters.

S5 is specifically: cut out each connected domain in the image T _out , perform KNN prediction in turn, and then compare the prediction result with the probability threshold pn, the probability threshold pn is obtained by multiple experiments, if the prediction probability is greater than pn, then the prediction result As the character value of the connected domain, otherwise it is determined that the connected domain character is defective.

Among them, KNN prediction is to calculate the distance between the test data and the database data, sort the distances from small to large, and select the k database data with the smallest distance. In this embodiment, k=1 is taken as the prediction result;

Example 2:

This embodiment is a detection system applying the method for detecting the production date label of a commodity described in Embodiment 1. As shown in FIG. 6 , the system includes a stand 8 , a conveyor belt 6 , a bracket 2 , an industrial camera 3 , a light source 5 , and a processor 1;

The conveyor belt is arranged on the stand 8; it is used to convey the commodity 7 to be tested;

One end of the bracket 2 is fixed to the side of the stand 8, and the other end is used to fix the industrial camera 3, so that the industrial camera is located above the conveyor belt 6, and the lens 4 of the industrial camera 3 faces downwards, which is used to capture the production date label of the goods 7 to be tested on the conveyor belt. image, and transmit the captured image information to the processor;

The light source 5 is arranged below the industrial camera 3 to supplement the light source for the industrial camera 3;

The processor 1 controls the industrial camera 3 to capture images, and performs data processing on the images captured by the industrial camera 3, including using HSV space and grayscale space to perform threshold segmentation on the target image, and filter out suspicious objects based on grayscale features and morphological features. Character connected domain.

The light source 5 is a ring light source.

The same or similar reference numbers correspond to the same or similar parts;

The terms describing the positional relationship in the accompanying drawings are only used for exemplary illustration, and should not be construed as a limitation on this patent;

Obviously, the above-mentioned embodiments of the present invention are only examples for clearly illustrating the present invention, rather than limiting the embodiments of the present invention. For those of ordinary skill in the art, changes or modifications in other different forms can also be made on the basis of the above description. There is no need and cannot be exhaustive of all implementations here. Any modification, equivalent replacement and improvement made within the spirit and principle of the present invention shall be included within the protection scope of the claims of the present invention.

Claims (8)

1. a commodity production date label detection method, is characterized in that, described method comprises: S1: Obtain the RGB image of the product surface, and preprocess the RGB image to obtain a binary image; S2: Perform connected domain screening on the binary image obtained by S1; obtain the connected domain set of the filtered suspicious targets; S3: perform rapid segmentation processing of the production date label with rotation on the connected domain set of the suspicious target, and obtain the image T _out of the production date character combination; S4: establish a character comparison library; obtain the images of the characters on the production date label through experiments, and form an image set CH={CH _i }, i=1, 2, 3,...,n; S5: Cut out each connected domain in the image T _out , and perform KNN prediction and recognition in turn; and obtain the prediction result. 2. commodity production date label detection method according to claim 1 is characterized in that, S1 comprises the following steps: S1.1: Obtain the RGB image O of the product surface and its grayscale image G; convert the image O to the HSV space to obtain the image H; perform a fixed threshold segmentation on the image G to obtain the binary image B, and the target pixel is set to 255, The background pixel is set to 0; S1.2: Let any pixel of the image H be Hi, and its HSV component is {h, _s , v _} ; let the pixel in the image B with the same pixel position as Hi be _Bi ; if _Hi and _Bi satisfy Condition Con, set B _i to 0, where Condition Con is: Among them, R _h , R _s , and R _v represent the range of H, S, and V component values of the target pixel; because the target object and the detection environment are fixed, the interval range of R _h , R _s , and R _v is obtained by multiple experiments; traverse the images H and Image B, obtain a new obtained binary image B. 3. commodity production date label detection method according to claim 2 is characterized in that, S2 comprises the following steps: S2.1: Detect all connected domains in the binary image B after preprocessing, and obtain a connected domain set Re={Re _i },i=1,2,3,...,n; S2.2: Set the connected domain subset of the suspicious target as Rc, and detect the smallest circumscribed rectangle of the connected domain Re _i , whose long side length is a and the short side length is b. If a∈R _a and b∈R _b , then Save Re _i in Rc, where R _a and R _b represent the interval range of the length of the long side and the length of the short side of the minimum circumscribed rectangle of the connected domain, respectively; S2.3: Traverse the connected domain set Re, and obtain the filtered connected domain subset Rc of suspicious targets. The filtered connected domain includes production date characters and some interference connected domains with similar shapes to production date characters. 4. commodity production date label detection method according to claim 3 is characterized in that, S3 comprises the following steps: S3.1: Calculate the center coordinates of the smallest circumscribed rectangle of each connected domain in the connected domain subset Rc, and obtain the center coordinate set C={C _i }, i=1, 2, 3,..., n, and represent the position of the corresponding element in Rc with the coordinate set; S3.2: Define the set N={N _i }, i=1, 2, 3, ......, n; let i=0, j=0; S3.3: Take the element C _i as the center, select a rectangular area Rect_w whose width and height are w, and count the number n of elements in the set C that the Rect_w area contains. If n>t, where t is the preset threshold, Then let N _j =C _i , j=j+1, otherwise start the next step; S3.4: Execute i=i+1; and judge whether C _i ∈ C is established, if so, go back to step S3.3, otherwise start the next step; S3.5: Establish a solution pool for selecting j elements from the set N according to the permutation and combination: P={P _k },k=1,2,3,...,n That is, the set P contains all the cases where j elements are selected from the set N, and each element P _k in the set P represents a situation where j elements are selected from the set N, and each element of the set P is different from each other. ; let k = 0; S3.6: adopt the scheme P _k to select j elements from the set N, form the j elements into a whole T, and calculate the ratio f of the length of the short side to the long side of the minimum circumscribed rectangle Rect_T of T; S3.7: Set the threshold range of the ratio interval to ratio, that is, the range of the ratio of the short side to the long side of the minimum circumscribed rectangle of the connected domain. The ratio interval is determined according to the character size. Since the detection target is fixed, its value can be obtained by multiple experiments. f∈ratio, then get the rotation angle of Rect_T Then start step S3.10, otherwise let k=k+1, and then start the next step; S3.8: If P _k ∈ P, go back to step S3.6, otherwise set j=j-1, and then start the next step; S3.9: If j>n _t , go back to step S3.5, where n _t is determined by the number of target characters, otherwise output segmentation failure information, and end the segmentation process; S3.10: Cut out the overall image of the elements in the Rect_T area in the connected domain set Rc, and according to the rotation angle Rotate it until the long side of the image coincides with the horizontal direction to obtain the image T _out of the combination of production date characters. 5. The method for detecting a product production date label according to claim 1, wherein the characters in S4 include 0-9, a-z, A-Z, ".", "/", "\" and "-". 6. The method for detecting the date of manufacture of goods according to any one of claims 1-5, wherein S5 is specifically: cutting out each connected domain in the image T _out , performing KNN prediction successively and then combining the prediction result with the result. The probability threshold pn is compared, and if the predicted probability is greater than pn, the predicted result is used as the character value of the connected domain, otherwise it is determined that the connected domain character is defective. 7. A detection system applying the method for detecting a product production date label according to any one of claims 1-6, wherein the system comprises a gantry, a conveyor belt, a bracket, an industrial camera, a light source, and a processor; The conveyor belt is set on the stand; it is used to convey the goods to be tested; One end of the bracket is fixed on the side of the stand, and the other end is used to fix the industrial camera, so that the industrial camera is located above the conveyor belt, and the industrial camera lens is facing down, which is used to capture the image of the production date label of the goods to be inspected conveyed on the conveyor belt, and will capture The image information is transmitted to the processor; The light source is arranged under the industrial camera to supplement the light source for the industrial camera; The processor controls the industrial camera to capture images, and performs data processing on the images captured by the industrial cameras, including using HSV space and grayscale space to perform threshold segmentation on the target image, and screening out suspicious character connected domains based on grayscale features and morphological features . 8. The product production date label detection system according to claim 7, wherein the light source is a ring light source.