Small box appearance detection and rejection control method based on multi-end cooperation

Abstract

The application discloses a kind of based on multi-end cooperation's little box appearance detection and elimination control method, the main design concept of the application is in, first inspection is carried out to cigarette little box using traditional machine vision detection device at winding equipment end, when detecting flaw, flaw product will be directly eliminated, simultaneously based on image compression algorithm, little box image is compressed and sent to edge end appearance re-inspection server, image sent by machine vision detection system of multiple equipment ends is secondly inspected using artificial intelligence algorithm based on deep learning at edge end;And since little box generally has only one elimination station, based on little box and strip box association mechanism, little box is positioned, so as to contain the little box with first undetected flaw, but re-inspected flaw is eliminated at strip box elimination station.This application can improve detection accuracy while reducing missed elimination rate and false elimination rate, further guarantee quality and reduce unnecessary detection loss.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of cigarette manufacturing, and particularly to a small box appearance detection and rejection control method based on multi-end cooperation. BACKGROUND

[0002] In the aspect of cigarette market consumption, with the improvement of economic development and consumption level, consumers' requirements and expectations for cigarette quality are increasingly high, and appearance quality is the most direct way for consumers to measure the manufacturing level of an enterprise. Due to the importance of appearance quality, the industry has always taken the introduction point of appearance defects in the manufacturing process as the direction, and carried out a lot of researches on the generation mechanism of various appearance defects. Through improving equipment components and structure, optimizing auxiliary material quality, and standardizing personnel operation, certain results have been achieved. However, due to the limitations of equipment inherent capacity, auxiliary material defects, and operation uncertainty, and the fact that accidental fluctuations are extremely difficult to eliminate, it is inevitable to produce flaws affecting consumer experience. Taking the cigarette packaging process as an example, usually due to auxiliary material reasons, foreign matter on the conveying belt, and loose mechanical components during the production and conveying process of cigarette packs, surface scratches, cigarette pack damage, and cigarette pack deformation may occur.

[0003] Therefore, cigarette factories in the tobacco industry have begun to gradually extend the research from the introduction point of defects to the control point. The traditional product appearance quality control mode is to detect by artificial visual inspection, and then feedback control after finding the problem. However, due to the gap between the sampling ratio and the production speed, the point of finding defects is seriously lagging behind. Therefore, the transformation of appearance quality control from sampling inspection to online visual flaw detection has become a common and urgent need of industrial enterprises.

[0004] Currently, visual inspection in cigarette production has essentially covered the entire production chain, from individual cigarettes to cartons. The fastest inspection speed is 3 milliseconds per scan, with an accuracy requirement of 2 square millimeters. A typical cigarette factory production workshop has several hundred sets of these systems. Regarding the overall visual inspection equipment, devices based on traditional algorithms (such as filtering, histograms, edge detection, image segmentation, morphological transformation, contour extraction, template matching, and geometric transformation) are widely used for online inspection of cigarette product appearance quality. However, due to the numerous interferences in the cigarette production environment and the randomness of product defect samples, some missed or false detections are unavoidable, posing a certain quality risk and causing unnecessary costs. Therefore, in recent years, industry professionals have begun to gradually research the application of deep learning technology to online inspection of cigarette product appearance quality. Deep learning offers high precision, but due to the extremely fast production speed of cigarettes, directly applying deep learning algorithms cannot meet the real-time requirements of detection speed. Moreover, the cost of deep learning-based detection devices is relatively high. Therefore, directly applying deep learning to the appearance inspection of cigarette products has always presented certain technical challenges, and problems such as insufficient accuracy, missed detections, and false detections remain to be solved. Summary of the Invention

[0005] In view of the above, the present invention aims to provide a method for small box appearance detection and rejection control based on multi-terminal collaboration, so as to solve the aforementioned technical problems.

[0006] The technical solution adopted in this invention is as follows:

[0007] This invention provides a method for small box appearance inspection and rejection control based on multi-terminal collaboration, including:

[0008] Several roll packaging units use a pre-configured equipment-side small box appearance inspection system to acquire and save small box images at the inspection station;

[0009] The first appearance defect detection is performed based on the acquired small box image and the preset machine vision algorithm;

[0010] Based on the results of the first inspection, the small boxes that are found to be defective are rejected at the small box rejection station.

[0011] The device-side small box appearance inspection system compresses the image of each small box and sends it to the edge-side appearance re-inspection server;

[0012] After decompressing the image of each small box, the edge-end appearance re-inspection server performs appearance defect re-inspection based on the preset AI intelligent algorithm;

[0013] The edge-end appearance re-inspection server feeds back the re-inspection result to the equipment end of the wrapping machine group, and the equipment end rejects the carton containing the small box with missed defects at a carton rejection station according to the association relationship between the small box and the carton.

[0014] In at least one possible implementation, the way in which the small box and the carton establish the association relationship includes: associating the small box and the carton through the positions of preset workstations of the wrapping machine group, or using an identification analysis strategy to associate the small box and the carton.

[0015] In at least one possible implementation, the control method further includes: the edge-end appearance re-inspection server performs cross-parameter optimization and / or detection point optimization on the equipment end of each wrapping machine group according to the re-inspection result.

[0016] In at least one possible implementation, the detection point optimization includes: when the edge-end appearance re-inspection server detects that there is a missed detection situation, issuing a detection point addition instruction to add a detection point to the small box appearance detection system at the corresponding equipment end.

[0017] In at least one possible implementation, the parameter optimization includes: when the edge-end appearance re-inspection server detects that there is a missed detection situation of a preset defect type, issuing a parameter modification instruction to make the small box appearance detection system at the corresponding equipment end change parameters for the preset defect type.

[0018] In at least one possible implementation, the small box vision appearance detection system is independently configured at the equipment end of each of the plurality of wrapping machine groups.

[0019] In at least one possible implementation, one edge-end appearance re-inspection server is centrally configured for the plurality of wrapping machine groups.

[0020] In at least one possible implementation, the small box appearance detection system at the equipment end compresses each small box image includes: compressing the complete image of each small box based on a preset deep learning algorithm compression strategy.

[0021] The main design concept of the present application is that a first inspection of cigarette packets is performed by a machine vision detection device based on a traditional algorithm at a packaging equipment end, and when a defect is detected, the defective product is directly rejected, and based on an image compression algorithm, the packet image is compressed and sent to an edge-end appearance re-inspection server, and a second inspection of the images sent by the machine vision detection system of multiple equipment ends is performed at the edge end using an artificial intelligence algorithm based on deep learning. Since the packet generally has only one rejection station, based on the association mechanism of the packet and the carton, the packet is positioned so that the packet containing the first undetected defect but the re-inspected defect is rejected at the carton rejection station. The present application can improve the detection accuracy while reducing the miss rejection rate and the false rejection rate, further ensuring the quality and reducing unnecessary detection loss. BRIEF DESCRIPTION OF DRAWINGS

[0022] To make the objectives, technical solutions, and advantages of the present application clearer, the present application will be further described below with reference to the drawings, in which:

[0023] Figure 1 A flowchart of the packet appearance detection and rejection control method based on multi-end cooperation provided by the embodiments of the present application is shown. DETAILED DESCRIPTION

[0024] The embodiments of the present application will be described in detail below, and examples of the embodiments are shown in the drawings, in which the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present application, and cannot be interpreted as a limitation on the present application.

[0025] The present application proposes an embodiment of a packet appearance detection and rejection control method based on multi-end cooperation, specifically as shown in Figure 1 , which includes:

[0026] Step S1: A plurality of cigarette packaging machines perform packet image acquisition and save at the detection station through a pre-configured equipment-end packet appearance detection system;

[0027] Step S2: Perform a first appearance defect detection according to the acquired packet image and in combination with a pre-set machine vision algorithm;

[0028] Step S3: According to the first detection result, the packets detected as defective are rejected at the packet rejection station;

[0029] Step S4: The equipment-end packet appearance detection system compresses each packet image and sends it to an edge-end appearance re-inspection server;

[0030] Step S5, after decompressing each small box image, the edge-end appearance re-inspection server performs appearance defect re-inspection based on a preset AI intelligent algorithm.

[0031] Step S6, the edge-end appearance re-inspection server feeds back the re-inspection result to the device end of the wrapping machine set, and the device end removes the wrapping box containing the defect small box missed in the re-inspection at a wrapping box removal station according to the association relationship between the small box and the wrapping box.

[0032] The association relationship between the small box and the wrapping box includes: associating the small box and the wrapping box through the position of the preset station of the wrapping machine set, or using an identification analysis technology to associate the small box and the wrapping box.

[0033] Further, the control method further comprises: the edge-end appearance re-inspection server cross-parameter optimization and / or detection point optimization of the device end of each wrapping machine set according to the re-inspection result.

[0034] Specifically, the detection point optimization includes: when the edge-end appearance re-inspection server detects that there is a missed inspection situation, issuing a detection point addition instruction to add a detection point to the small box appearance detection system of the corresponding device end.

[0035] The parameter optimization includes: when the edge-end appearance re-inspection server detects that there is a missed inspection situation of a preset defect type, issuing a parameter modification instruction to change the parameters of the small box appearance detection system of the corresponding device end for the preset defect type.

[0036] Regarding the device-end small box appearance detection system involved in the control method, specifically, the wrapping machine sets are independently configured with small box visual appearance detection systems at the device end, that is, each of the multiple machine sets has a device-end small box appearance detection system that performs its own function.

[0037] Regarding the edge-end appearance re-inspection server involved in the control method, preferably, the wrapping machine sets are centrally configured with one edge-end appearance re-inspection server, that is, the multiple machine sets share one re-inspection server.

[0038] Finally, it can be further supplemented that the device-end small box appearance detection system compresses each small box image, including: compressing the complete image of each small box based on a preset deep learning algorithm compression strategy.

[0039] With the above embodiments, based on a certain hard box hard strip packaging machine group, the packaging machine is generally divided into a small box packaging machine and a strip box packaging machine. A small box appearance quality machine vision detection system based on a traditional visual detection algorithm and a small box rejection port corresponding to the small box packaging machine are arranged in the small box packaging machine. A strip box rejection port is arranged in the strip box packaging machine. The packaging machine group is connected to an edge visual detection server based on an AI visual detection algorithm. The specific implementation steps are as follows:

[0040] A traditional machine vision detection algorithm is executed by the device end small box appearance detection system of each machine group. Small box image acquisition and first appearance defect detection based on a traditional machine vision algorithm are performed at the detection station. According to the first detection result, the small boxes with defects are rejected at the small box rejection port. This is the first inspection. The small box images are completely saved. In actual operation, the pictures can be saved in JPG or BMP format.

[0041] Each picture retained by the device end appearance detection system of multiple machine groups is compressed based on a deep learning algorithm (the specific method can refer to existing solutions in the industry, such as ZL202110379367.9, etc.). After compression, the file size of the image can be reduced to one tenth of the original, thereby facilitating real-time and efficient transmission to the edge visual re-inspection server and not occupying too much server space.

[0042] The edge appearance re-inspection server decompresses each small box image sent by the machine group and performs appearance defect detection based on AI intelligent algorithm again (there are a large number of technical solutions that can be referred to in the industry for packaging appearance detection through artificial intelligence algorithm, which will not be described here). This process completes the re-inspection of the small box image.

[0043] The edge appearance re-inspection server feeds back the re-inspection result to the device end. When the re-inspection is normal, it is released. When the re-inspection is abnormal, since the detection accuracy of the edge end is higher than that of the traditional algorithm, when the re-inspection finds a small box with defects, according to the running speed of the machine group, the small box missed by the traditional algorithm has a high probability of missing the small box rejection port. Therefore, the present application proposes that the association between the small box and the strip box needs to be established at this stage (specifically, the packaging machine group has a strict correspondence relationship between the stations, so the stations can be associated; or, identification analysis technology can also be used for association) to locate the missed small box in which strip box, and then the strip box containing the small box with defects is rejected at the strip box rejection port.

[0044] Further, the edge end appearance re-inspection server can also realize cross parameter optimization and detection point (range) optimization of the small box appearance detection system at each device end of the multiple machine group devices producing the same brand, iteratively improve the detection accuracy and coverage range of the visual detection system at the device end, improve the detection accuracy, and reduce the missed detection rate and the false detection rate. For example: (1) each device end small box appearance detection system is provided with corresponding detection points, when the edge end appearance re-inspection server detects that there is a missed detection, the detection point addition instruction can be issued to add detection points at the corresponding device end small box appearance detection system, so as to improve the detection accuracy and reduce the missed detection rate; (2) when the edge end appearance re-inspection server finds that there is a missed detection of a certain type of defect, such as a missed detection caused by the reflectivity of a certain packaging material or the change of the light condition, the edge end appearance re-inspection server can issue a parameter modification instruction to reduce the sensitivity of certain detection points, thereby reducing the false detection rate.

[0045] In order to verify the above-mentioned scheme of the present application, the following comparative test is carried out.

[0046] Comparative machine group: two groups of ZB45 packaging equipment, one of which uses the detection method proposed by the present application, and the other is a control group which does not use the aforementioned method of the present application.

[0047] Product specification: 84*7.7 regular cigarette, hard box and hard strip machine group;

[0048] Comparison method: because the data volume is too large, only ten thousand pictures saved are continuously viewed for manual identification, the detection accuracy, missed detection rate and false detection rate of the small box appearance quality online detection are compared respectively, and the defects are combined in the way of manual production and natural generation;

[0049] Among them:

[0050]

[0051]

[0052]

[0053]

[0054] It can be seen that the main technical performance index of the small box appearance quality detection of the cigarette making machine group using the method of the present application is significantly better than that of the control machine group which does not use the method of the present application, and in actual application, multiple machine group devices can only be configured with one server based on intelligent algorithms such as deep learning, and there is no too strict requirement for the running speed of the machine group, in the subsequent carton rejection stage, only the small boxes with defects in the carton need to be taken out, the rest can still be recycled and reused, and there is no excess consumption. Overall, the present application has good practical application significance.

[0055] In summary, the main design concept of the present application is that a first inspection of cigarette packets is performed by a machine vision detection device based on a traditional algorithm at the end of a packaging device, and when a defect is detected, the defective product is directly rejected, and based on an image compression algorithm, the packet image is compressed and sent to an edge appearance re-inspection server, and a second inspection of the images sent by the machine vision detection system of multiple device ends is performed at the edge using an artificial intelligence algorithm based on deep learning. Since the packet generally has only one rejection station, the packet is positioned based on the association mechanism of the packet and the carton to allow the packet containing a defect that is not detected in the first inspection but is re-inspected to be rejected at the carton rejection station.

[0056] In the embodiments of the present application, "at least one" means one or more, and "multiple" means two or more. The "and / or" describes the association relationship of the associated objects, which means that there can be three kinds of relationships, for example, A and / or B, which can represent the cases of A alone, A and B together, and B alone. Wherein A and B can be singular or plural. The character " / " generally represents an "or" relationship between the front and rear associated objects. "At least one of the following" and the like means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, wherein a, b, and c can be single or multiple.

[0057] The above embodiments according to the drawings illustrate the structure, features and effects of the present application, but the above is only a preferred embodiment of the present application, and it should be noted that the technical features involved in the above embodiments and preferred modes can be reasonably combined and matched into various equivalent schemes by those skilled in the art without departing from or changing the design idea and technical effects of the present application. Therefore, the present application is not limited by the drawings shown, and any changes or modifications made in accordance with the concept of the present application, or equivalent embodiments with equivalent changes, shall be within the scope of protection of the present application.

Claims

1. A method for small box appearance inspection and rejection control based on multi-terminal collaboration, characterized in that, include: Several roll packaging units use a pre-configured equipment-side small box appearance inspection system to acquire and save small box images at the inspection station; The first appearance defect detection is performed based on the acquired small box image and the preset machine vision algorithm; Based on the results of the first inspection, the small boxes that are found to be defective are rejected at the small box rejection station. The device-side small box appearance inspection system compresses the image of each small box and sends it to the edge-side appearance re-inspection server; After decompressing the image of each small box, the edge-end appearance re-inspection server performs appearance defect re-inspection based on the preset AI intelligent algorithm; The edge-end appearance re-inspection server feeds back the re-inspection results to the roll packaging unit equipment. Based on the relationship between the small boxes and the strip boxes, the equipment rejects the strip boxes containing the defective small boxes that were missed in the inspection at the strip box rejection station.

2. The method for small box appearance inspection and rejection control based on multi-terminal collaboration according to claim 1, characterized in that, The ways to establish a relationship between small boxes and cartons include: associating small boxes and cartons through the preset position of the wrapping machine unit, or using an identifier resolution strategy to associate small boxes and cartons.

3. The method for small box appearance inspection and rejection control based on multi-terminal collaboration according to claim 1, characterized in that, The control method further includes: the edge-end appearance re-inspection server performs cross-parameter optimization and / or detection point optimization on the equipment end of each roll-packing unit based on the re-inspection results.

4. The method for small box appearance inspection and rejection control based on multi-terminal collaboration according to claim 3, characterized in that, The optimization of the detection points includes: when the edge-end appearance re-inspection server detects a missed detection, it issues a command to add a new detection point, and adds a new detection point in the corresponding device-end small box appearance inspection system.

5. The method for small box appearance inspection and rejection control based on multi-terminal collaboration according to claim 3, characterized in that, The parameter optimization includes: when the edge-end appearance re-inspection server detects a missed detection of a preset defect type, it issues a parameter modification instruction, causing the corresponding device-end small box appearance inspection system to change the parameters for the preset defect type.

6. The method for small box appearance inspection and rejection control based on multi-terminal collaboration according to claim 1, characterized in that, Several roll packaging units are each independently equipped with a small box visual appearance inspection system at the equipment end.

7. The method for small box appearance inspection and rejection control based on multi-terminal collaboration according to claim 1, characterized in that, Several roll packaging machines are centrally configured with one edge-end appearance re-inspection server.

8. The method for small box appearance inspection and rejection control based on multi-terminal collaboration according to any one of claims 1 to 7, characterized in that, The device-side small box appearance detection system compresses each small box image by: compressing the complete image of each small box based on a preset deep learning algorithm compression strategy.

Images