Cigarette package appearance quality inspection device and method, cigarette package packaging machine

Abstract

The present disclosure provides a cigarette pack appearance quality inspection device and method, and a cigarette packing machine. The cigarette pack appearance quality inspection device includes: a first positioning sensor; a second positioning sensor; a controller, which is used to issue an image acquisition instruction after receiving information from the first positioning sensor that the cigarette pack is transported to the first position, and detect whether the cigarette pack is set with rejection mark information after receiving information from the second positioning sensor that the cigarette pack is transported to the second position, and issue a rejection instruction when the cigarette pack is set with rejection mark information; an image acquisition device, which is used to acquire image information of the cigarette pack after receiving the image acquisition instruction; an artificial intelligence processing unit, which is used to detect the image information of the cigarette pack using an image detection model, and set the rejection mark information for the cigarette pack when it is detected that the appearance quality of the cigarette pack is defective; and a rejection device, which is used to perform a rejection operation on the cigarette pack after receiving the rejection instruction.

Description

Technical Field

[0001] The present disclosure relates to the technical field of tobacco manufacturing, and in particular to a cigarette pack appearance quality inspection device and method, and a cigarette pack packaging machine. Background Art

[0002] In cigarette packaging production technology, the traditional cigarette packaging appearance inspection system can detect quality defects in the preset area of ​​the packaging appearance. The two core algorithms of such cigarette packaging appearance inspection systems are color matching and standard template matching. Both algorithms will grayscale the image, converting the original RGB (red, green and blue) three-channel color values ​​into grayscale images, reducing the three-dimensional image to a one-dimensional image.

[0003] Color matching is to calculate the mean and mean variance of the pixels in the detection area, and then compare the two indicators respectively; template matching will record the pixel value array in the detection area, and then compare the similarity of the array. During real-time detection, set the standard mean, mean variance and template pattern pixel value for the color matching and template matching algorithm tools respectively, and compare the image to be detected with the standard value or standard template. In the color matching tool, if one of the mean and mean variance of the standard to be detected is greater than the standard value, the detection fails; if all are less than the standard value, the detection passes. In the template matching tool, if the pixel value score of the area to be detected is less than the pixel value score of the standard template, the detection fails; if the score is greater than the standard score, the detection passes.

[0004] However, the detection accuracy of traditional algorithms is mainly affected by several aspects: First, after the grayscale dimension reduction processing, the traditional algorithm has a large information loss in the detection area. Second, statistical parameters such as color mean and mean variance can be used to count the appearance of pure colors, but the accuracy is not high for detection areas with large color changes. Third, traditional detection frames are usually set at the edge of the cigarette package (an area prone to defects). In actual use, the cigarette package will shake irregularly on the conveyor belt, causing the detection frame to easily produce false detections. Fourth, traditional algorithms usually use many detection frames for combined detection. Any inaccurate detection of any detection frame will affect the final result, and the detection stability is poor. Therefore, the traditional cigarette package appearance detection system has the problem of low equipment detection accuracy. Summary of the invention

[0005] A technical problem solved by the present disclosure is to provide a cigarette package appearance quality detection device to improve the detection accuracy of the cigarette package appearance quality.

[0006] According to one aspect of the present disclosure, a device for inspecting the appearance quality of cigarette packages is provided, comprising: a first positioning sensor, for detecting the cigarette package when the cigarette package is transported to a first position, and sending information that the cigarette package is transported to the first position to a controller; a second positioning sensor, for detecting the cigarette package when the cigarette package is transported to a second position, and sending information that the cigarette package is transported to the second position to the controller, wherein the cigarette package is transported from the first position to the second position by a conveyor belt; and the controller, for issuing an image acquisition instruction after receiving the information that the cigarette package is transported to the first position, and transmitting an image acquisition instruction after receiving the information that the cigarette package is transported to the second position. After receiving the information set, it is detected whether the cigarette package is set with rejection mark information, and a rejection instruction is issued when the cigarette package is set with rejection mark information; an image acquisition device is used to acquire the image information of the cigarette package after receiving the image acquisition instruction; an artificial intelligence processing unit is used to receive the image information of the cigarette package from the image acquisition device, detect the image information of the cigarette package by using an image detection model, and set rejection mark information for the cigarette package when it is detected that there are defects in the appearance quality of the cigarette package, and send the rejection mark information to the controller; and a rejection device is used to perform a rejection operation on the cigarette package after receiving the rejection instruction from the controller.

[0007] In some embodiments, the artificial intelligence processing unit is also used to train the image detection model using image information of cigarette packages with known appearance quality.

[0008] In some embodiments, the artificial intelligence processing unit is also used to not set rejection mark information for the cigarette package when it is detected that there is no defect in the appearance quality of the cigarette package; the controller is also used to not issue a rejection instruction when no rejection mark information is set for the cigarette package.

[0009] In some embodiments, the image acquisition device includes: a first camera, used to photograph the first side and the second side of the cigarette package to obtain image information of the first side and the second side, and send the image information of the first side and the second side to the artificial intelligence processing unit, wherein the first side is adjacent to the second side; a second camera, used to photograph the third side and the fourth side of the cigarette package to obtain image information of the third side and the fourth side, and send the image information of the third side and the fourth side to the artificial intelligence processing unit, wherein the third side is opposite to the first side, and the fourth side is opposite to the second side; and a third camera, used to photograph the top surface of the cigarette package to obtain image information of the top surface, and send the image information of the top surface to the artificial intelligence processing unit.

[0010] In some embodiments, the cigarette pack appearance quality inspection device further includes: a light source for illuminating the cigarette pack after receiving a lighting instruction from the controller; wherein the controller is also used to issue the lighting instruction after receiving information that the cigarette pack is transported to the first position.

[0011] In some embodiments, the first positioning sensor is a detection position fiber optic sensor, and the first position is a detection position; the second positioning sensor is a rejection position fiber optic sensor, and the second position is a rejection position; the controller includes an input and output control board; and the artificial intelligence processing unit includes an industrial computer or a server.

[0012] In some embodiments, the cigarette pack appearance quality detection device also includes: an auxiliary positioning sensor, arranged next to the conveyor belt and between the first positioning sensor and the second positioning sensor, for triggering a position signal when detecting a cigarette pack transported to the corresponding position, and transmitting the position signal to the controller; wherein the controller is used to obtain the number n of cigarette packs located before the cigarette pack with defective appearance quality according to the received position signal, and after receiving the information sent by the second positioning sensor that the n+1th cigarette pack is transported to the second position, issue a rejection instruction if the cigarette pack is set with rejection mark information, wherein n is an integer greater than or equal to 0.

[0013] In some embodiments, the cigarette pack appearance quality inspection device further includes: a display for displaying image information of the cigarette pack and statistical information of cigarette pack rejection; wherein the artificial intelligence processing unit is also used to collect statistics on the cigarette pack rejection to obtain the statistical information, and send the statistical information to the display.

[0014] In some embodiments, the cigarette package appearance quality detection device also includes: a switch for transmitting data; wherein the controller is communicatively connected to the image acquisition device via the switch, the image acquisition device is communicatively connected to the artificial intelligence processing unit and the display respectively via the switch, and the artificial intelligence processing unit is communicatively connected to the controller and the display respectively via the switch.

[0015] In some embodiments, the image detection model corresponds to the brand of the cigarette package; the image detection model is a convolutional neural network.

[0016] According to another aspect of the present disclosure, a method for detecting the appearance quality of a cigarette package is provided, comprising: when a cigarette package is transported to a first position, a first positioning sensor detects the cigarette package and sends information that the cigarette package is transported to the first position to a controller; after receiving the information that the cigarette package is transported to the first position, the controller sends an image acquisition instruction to an image acquisition device; after receiving the image acquisition instruction, the image acquisition device acquires image information of the cigarette package and sends the image information of the cigarette package to an artificial intelligence processing unit; the artificial intelligence processing unit detects the image information of the cigarette package using an image detection model, and upon detecting that there is a problem with the appearance quality of the cigarette package In the event of a defect, a rejection mark information is set for the cigarette package and the rejection mark information is sent to the controller; when the cigarette package is transported to the second position, the second positioning sensor detects the cigarette package and sends the information that the cigarette package is transported to the second position to the controller, wherein the cigarette package is transported from the first position to the second position by the conveyor belt; after receiving the information that the cigarette package is transported to the second position, the controller detects whether the rejection mark information is set for the cigarette package, and sends a rejection instruction to the rejection device if the rejection mark information is set for the cigarette package; and the rejection device performs a rejection operation on the cigarette package after receiving the rejection instruction.

[0017] In some embodiments, the method for detecting the appearance quality of cigarette packages further includes: training the image detection model using image information of cigarette packages with known appearance quality, wherein the image detection model corresponds to the brand of the cigarette packages.

[0018] In some embodiments, the step of training the image detection model includes: when there is no corresponding image detection model for the brand of the cigarette package, turning off the rejection device, establishing the brand of the cigarette package and loading the brand; adjusting the parameters of the image acquisition device so that the image acquisition device is in the best state for imaging the cigarette package; and using the image acquisition device to capture image information of the cigarette package and input it into the artificial intelligence processing unit to train the image detection model.

[0019] In some embodiments, the step of training the image detection model includes: when there is a corresponding image detection model for the brand of cigarette pack, using the image acquisition device to collect image information of the cigarette pack and input it into the artificial intelligence processing unit to optimize the image detection model.

[0020] In some embodiments, the cigarette package appearance quality inspection method further includes: after receiving the information that the cigarette package is transported to the first position, the controller sends a lighting instruction to the light source so that the light source illuminates the cigarette package.

[0021] According to another aspect of the present disclosure, a device for detecting appearance quality of cigarette packages is provided, comprising: a memory; and a processor coupled to the memory, wherein the processor is configured to execute the aforementioned method based on instructions stored in the memory.

[0022] According to another aspect of the present disclosure, a cigarette packing machine is provided, comprising: the cigarette pack appearance quality inspection device as described above.

[0023] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which computer program instructions are stored. When the instructions are executed by a processor, the method described above is implemented.

[0024] The above-mentioned cigarette package appearance quality detection device improves the detection accuracy of cigarette package appearance quality.

[0025] Other features and advantages of the present disclosure will become apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings, which constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0027] The present disclosure may be more clearly understood from the following detailed description with reference to the accompanying drawings, in which:

[0028] Figure 1 is a schematic diagram showing the structure of a cigarette pack appearance quality inspection device according to some embodiments of the present disclosure;

[0029] Figure 2 is a schematic diagram showing the structure of a cigarette pack appearance quality inspection device according to other embodiments of the present disclosure;

[0030] Figure 3A is a schematic diagram showing a first camera and a second camera taking a photo of a cigarette pack according to some embodiments of the present disclosure;

[0031] Figure 3B is a schematic diagram showing a third camera taking a photo of a cigarette pack according to some embodiments of the present disclosure;

[0032] Figure 4 is a schematic diagram showing artificial intelligence deep learning according to some embodiments of the present disclosure;

[0033] Figure 5 is a schematic diagram illustrating feature extraction by deep learning according to some embodiments of the present disclosure;

[0034] Figure 6 is a schematic diagram showing the structure of a convolutional neural network according to some embodiments of the present disclosure;

[0035] Figure 7 is a schematic diagram showing a model of a convolutional neural network according to some embodiments of the present disclosure;

[0036] Figure 8 is a schematic diagram showing that data according to some embodiments of the present disclosure converges quickly after passing through a convolutional layer and a hidden layer;

[0037] Fig. 9 is a schematic diagram illustrating a convolution operation according to some embodiments of the present disclosure;

[0038] Fig.10 is a schematic diagram illustrating a sampling operation according to some embodiments of the present disclosure;

[0039] Fig.11 is a schematic diagram illustrating a local field technique according to some embodiments of the present disclosure;

[0040] Fig.12 is a flow chart showing a method for detecting the appearance quality of cigarette packages according to some embodiments of the present disclosure;

[0041] Fig.13 is a schematic diagram showing the structure of a cigarette pack appearance quality inspection device according to other embodiments of the present disclosure;

[0042] Fig.14 2 is a schematic diagram showing the structure of a device for inspecting the appearance quality of cigarette packages according to other embodiments of the present disclosure. DETAILED DESCRIPTION

[0043] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present disclosure unless otherwise specifically stated.

[0044] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0045] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present disclosure, its application, or uses.

[0046] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0047] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0048] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0049] Figure 1 is a schematic diagram showing the structure of a cigarette pack appearance quality inspection device according to some embodiments of the present disclosure. Figure 1 As shown, the cigarette package appearance quality inspection device includes: a first positioning sensor 110, a second positioning sensor 120, a controller 130, an image acquisition device 140, an artificial intelligence (AI) processing unit 150 and a rejection device 160.

[0050] The first positioning sensor 110 is arranged beside the conveyor belt and at a position corresponding to the first position of the cigarette pack ( Figure 1 The first positioning sensor 110 is electrically connected to the controller 130. The first positioning sensor 110 is used to detect the cigarette package when the cigarette package is transported to the first position, and send the information that the cigarette package is transported to the first position to the controller 130. For example, the first positioning sensor 110 is a detection position optical fiber sensor, and the first position is the detection position.

[0051] The second positioning sensor 120 is arranged beside the conveyor belt and at a position corresponding to the second position of the cigarette pack ( Figure 1 1 . The second positioning sensor 120 is electrically connected to the controller 130. The second positioning sensor 120 is used to detect the cigarette pack when it is transported to the second position, and send information that the cigarette pack is transported to the second position to the controller 130. Here, the cigarette pack is transported from the first position to the second position by the conveyor belt. For example, the second positioning sensor 120 is a rejection position optical fiber sensor, and the second position is the rejection position.

[0052] The controller 130 is used to issue an image acquisition instruction after receiving information that the cigarette package is transported to the first position, detect whether the cigarette package is set with rejection mark information after receiving information that the cigarette package is transported to the second position, and issue a rejection instruction when the cigarette package is set with rejection mark information. For example, the controller 130 may include an input and output control board (I / O control board).

[0053] The controller 130 may also be configured to not issue a rejection instruction when no rejection mark information is set for the cigarette package.

[0054] The image acquisition device 140 is disposed beside the conveyor belt and is electrically connected to the controller 130. The image acquisition device 140 is used to acquire image information of the cigarette pack after receiving an image acquisition instruction.

[0055] The artificial intelligence processing unit 150 is electrically connected to the controller 130 and the image acquisition device 140. The artificial intelligence processing unit 150 is used to receive image information of the cigarette package from the image acquisition device 140, detect the image information of the cigarette package using the image detection model, and set rejection mark information for the cigarette package when it is detected that the appearance quality (also referred to as packaging quality) of the cigarette package is defective, and send the rejection mark information to the controller 130. For example, the artificial intelligence processing unit 150 may include an industrial computer or a server.

[0056] In some embodiments, the image detection model corresponds to the brand of the cigarette pack. Here, the brand of the cigarette pack may include the pattern and color of the cigarette pack appearance, etc. Each brand corresponds to an image detection model, so that the appearance quality of cigarette packs of different brands can be detected separately.

[0057] In some embodiments, the image detection model is a convolutional neural network. For example, the image detection model can use a convolutional neural network known to those skilled in the art for detecting patterns, etc. In the present disclosure, such an image detection model is applied to the detection of the appearance quality of cigarette packages.

[0058] In other embodiments, the artificial intelligence processing unit is also used to not set rejection mark information for the cigarette pack when it is detected that there is no defect in the appearance quality of the cigarette pack.

[0059] The rejecting device 160 is arranged beside the conveyor belt. The rejecting device 160 is electrically connected to the controller 130. The rejecting device 160 is used to perform a rejecting operation on the cigarette pack after receiving a rejecting instruction from the controller 130. For example, the rejecting device 160 can be an air jet rejecting device.

[0060] It should be noted that the "electrical connection" described in the present disclosure may be a direct or indirect communication connection. For example, when describing that a specific device is electrically connected to other devices, the specific device may be directly connected to the other device without an intermediary device, or may not be directly connected to the other device but have an intermediary device.

[0061] So far, a cigarette pack appearance quality detection device according to some embodiments of the present disclosure is provided. The cigarette pack appearance quality detection device includes: a first positioning sensor, a second positioning sensor, a controller, an image acquisition device, an artificial intelligence processing unit and a rejection device. In the cigarette package appearance quality detection device, when the cigarette package is transported to the first position, the first positioning sensor detects the cigarette package and sends the information that the cigarette package is transported to the first position to the controller; after receiving the information that the cigarette package is transported to the first position, the controller sends an image acquisition instruction to the image acquisition device; after receiving the image acquisition instruction, the image acquisition device acquires the image information of the cigarette package and sends the image information of the cigarette package to the artificial intelligence processing unit; the artificial intelligence processing unit detects the image information of the cigarette package using the image detection model, and sets the rejection mark information for the cigarette package when it is detected that the appearance quality of the cigarette package is defective, and sends the rejection mark information to the controller; when the cigarette package is transported to the second position, the second positioning sensor detects the cigarette package and sends the information that the cigarette package is transported to the second position to the controller; after receiving the information that the cigarette package is transported to the second position, the controller detects whether the rejection mark information is set for the cigarette package, and sends the rejection instruction to the rejection device when the rejection mark information is set for the cigarette package; after receiving the rejection instruction, the rejection device performs a rejection operation on the cigarette package. The cigarette package appearance quality detection device detects the appearance quality of the cigarette package by using the image detection model, thereby improving the detection accuracy of the appearance quality of the cigarette package. For example, the study found that the defect detection accuracy of the cigarette pack appearance quality inspection device disclosed in the present invention can reach 99.5%. Moreover, the cigarette pack appearance quality inspection device disclosed in the present invention can reduce the false detection rate of qualified cigarette packs. For example, the study found that the false detection rate of qualified cigarette packs of the cigarette pack appearance quality inspection device is less than 0.02%.

[0062] In some embodiments, the artificial intelligence processing unit 150 can also be used to train the image detection model using image information of cigarette packages with known appearance quality.

[0063] For example, when there is no corresponding image detection model for the brand of cigarette packs, turn off the rejection device, establish the brand of cigarette packs and load the brand; adjust the parameters of the image acquisition device so that the image acquisition device is in the best state for imaging the cigarette packs; and use the image acquisition device to collect image information of the cigarette packs and input it into the artificial intelligence processing unit to train the image detection model.

[0064] For another example, when there is a corresponding image detection model for the brand of cigarette packages, image information of the cigarette packages is collected using an image acquisition device and input into an artificial intelligence processing unit to optimize the image detection model.

[0065] Figure 2 is a schematic diagram showing the structure of a cigarette pack appearance quality inspection device according to other embodiments of the present disclosure.

[0066] like Figure 2 As shown, the cigarette package appearance quality detection device includes a first positioning sensor 110, a second positioning sensor 120, a controller 130, an image acquisition device 140, an artificial intelligence processing unit 150 and a rejection device 160. In addition, Figure 2 Also shown are a conveyor belt 270 and cigarette packets 280.

[0067] like Figure 2 As shown, the first positioning sensor 110 is arranged beside the conveyor belt 270 and at a position corresponding to the first position of the cigarette pack 280. For example, the first positioning sensor 110 is arranged near the conveyor belt 270 and at a position corresponding to the first position of the cigarette pack 280. The second positioning sensor 120 is arranged beside the conveyor belt 270 and at a position corresponding to the second position of the cigarette pack 280. For example, the second positioning sensor 120 is arranged near the conveyor belt 270 and at a position corresponding to the second position of the cigarette pack 280. The image acquisition device 140 is arranged beside the conveyor belt 270. The rejection device 160 is arranged beside the conveyor belt 270.

[0068] In some embodiments, Figure 2 As shown, the image acquisition device 140 may include a first camera 141, a second camera 142, and a third camera 143. The first camera 141, the second camera 142, and the third camera 143 are arranged around the conveyor belt 270. The first camera 141, the second camera 142, and the third camera 143 are electrically connected to the controller 130, respectively. In addition, the first camera 141, the second camera 142, and the third camera 143 are electrically connected to the artificial intelligence processing unit 150, respectively. For example, the first camera 141, the second camera 142, and the third camera 143 are CCD (Charge-coupled Device) industrial cameras, respectively.

[0069] Figure 3A is a schematic diagram showing a first camera and a second camera taking a photo of a cigarette pack according to some embodiments of the present disclosure.

[0070] like Figure 3A As shown, the first camera 141 is used to take pictures of the first side 281 and the second side 282 of the cigarette pack 280 to obtain image information of the first side 281 and the second side 282, and send the image information of the first side 281 and the second side 282 to the artificial intelligence processing unit 150. The first side 281 is adjacent to the second side 282.

[0071] like Figure 3AAs shown, the second camera 142 is used to take pictures of the third side 283 and the fourth side 284 of the cigarette package 280 to obtain image information of the third side 283 and the fourth side 284, and send the image information of the third side 283 and the fourth side 284 to the artificial intelligence processing unit 150. The third side 283 is opposite to the first side 281, and the fourth side 284 is opposite to the second side 282. The third side 283 is adjacent to the fourth side 284.

[0072] Figure 3B is a schematic diagram showing a third camera taking a photo of a cigarette pack according to some embodiments of the present disclosure.

[0073] The third camera 143 is used to take a picture of the top surface 285 of the cigarette package 280 to obtain image information of the top surface 285, and send the image information of the top surface 285 to the artificial intelligence processing unit 150. In some embodiments, the third camera 143 can also take a picture of one side surface of the cigarette package 280 (for example, the first side surface 281, the second side surface 282, the third side surface 283 or the fourth side surface 284).

[0074] In the above embodiment, image information of five surfaces of the cigarette pack can be obtained by three cameras (ie, the first camera 141 , the second camera 142 and the third camera 143 ).

[0075] Back to Figure 2 In some embodiments, the cigarette pack appearance quality detection device may further include a light source 210. The light source 210 is disposed next to the conveyor belt 270. The light source 210 is electrically connected to the controller 130. The light source 210 is used to illuminate the cigarette pack 280 after receiving a lighting instruction from the controller 130. Here, the controller 130 may also be used to issue a lighting instruction after receiving information that the cigarette pack 280 is transported to the first position. For example, the light source includes an LED (Light Emitting Diode) lighting device. By illuminating the cigarette pack, it is convenient for the image acquisition device to collect the appearance image of the cigarette pack.

[0076] In some embodiments, Figure 2As shown, the cigarette pack appearance quality detection device may further include an auxiliary positioning sensor 220. The auxiliary positioning sensor 220 is arranged beside the conveyor belt 270, and the auxiliary positioning sensor 220 is between the first positioning sensor 110 and the second positioning sensor 120. The auxiliary positioning sensor 220 is electrically connected to the controller 130. The auxiliary positioning sensor 220 is used to trigger a position signal when detecting a cigarette pack transported to the corresponding position, and transmit the position signal to the controller 130. The controller 130 may be used to obtain the number n of cigarette packs located before the cigarette pack with defective appearance quality according to the received position signal, and after receiving the information sent by the second positioning sensor 120 that the n+1th cigarette pack is transported to the second position, issue a rejection instruction when the cigarette pack is set with a rejection mark information, wherein n is an integer greater than or equal to 0. Here, the auxiliary positioning sensor signal may play a shifting role, so as to facilitate the accurate rejection of unqualified cigarette packs without causing the problem of false rejection. For example, the auxiliary positioning sensor may be a positioning optical fiber sensor. The auxiliary positioning sensor can locate the position of the cigarette pack that is detected by the system to be defective and is about to be rejected.

[0077] For example, there is a cigarette pack without quality defects in front of the cigarette pack with defective appearance quality. The auxiliary positioning sensor 220 triggers a position signal when detecting the cigarette pack without quality defects. The controller learns from the position signal that the number of cigarette packs before the cigarette pack with defective appearance quality is 1, and the cigarette pack with defective appearance quality is the second cigarette pack. In this way, after receiving the information sent by the second positioning sensor that the second cigarette pack is transported to the second position, the controller determines that the current cigarette pack is the second cigarette pack, and issues a rejection instruction when the rejection mark information is set for the second cigarette pack, thereby rejecting the defective cigarette pack. Therefore, the auxiliary positioning sensor improves the rejection accuracy.

[0078] For another example, when two defective cigarette packs pass through the equipment successively, the auxiliary positioning sensor can accurately locate the positions of the two cigarette packs to prevent problems such as missed or inaccurate rejection.

[0079] In some embodiments, Figure 2 As shown, the cigarette pack appearance quality inspection device may further include a display 230. The display 230 is electrically connected to the image acquisition device 140 and the artificial intelligence processing unit 150. The display 230 may be used to display image information of cigarette packs and statistical information of cigarette pack rejection. Here, the artificial intelligence processing unit 150 may also be used to collect statistics on the situation of cigarette pack rejection to obtain statistical information, and send the statistical information to the display 230. By displaying the image information of cigarette packs and the statistical information of cigarette pack rejection, it is convenient for staff to obtain such information.

[0080] In some embodiments, the display 230 can host client software to provide functions such as real-time and historical detection image viewing, model training operations, and system configuration.

[0081] In some embodiments, Figure 2 As shown, the cigarette package appearance quality detection device may further include a switch 240. The switch is used to transmit data. The controller 130 is connected to the image acquisition device 140 through the switch 240. The image acquisition device 140 is connected to the artificial intelligence processing unit 150 and the display 230 through the switch 240. The artificial intelligence processing unit 150 is connected to the controller 130 and the display 230 through the switch 240. For example, the switch may be a 10 Gigabit switch. By setting the switch, the transmission of information such as data can be realized.

[0082] In some embodiments, the first positioning sensor 110, the second positioning sensor 120, the rejection device 160, the auxiliary positioning sensor 220 and the switch 240 are respectively electrically connected to the controller 130 through a field industrial bus, the image acquisition device 140 and the display 230 are respectively electrically connected to the switch 240 through a GigE (Gigabit Network) cable, and the artificial intelligence processing unit 150 is electrically connected to the switch 240 through a 10 Gigabit optical fiber.

[0083] The inventors of the present disclosure have discovered that, in addition to the problem of low equipment detection accuracy, the traditional cigarette package appearance detection system also has problems such as limited detection area and equipment maintenance.

[0084] Specifically, when the traditional detection algorithm is applied in practice, it is necessary to set a local detection area based on human experience, and the part outside the detection area cannot be detected. For the regular statistics of color values ​​in the local area, the detection area and parameter values ​​need to be preset within a reasonable range. Too large or too small detection areas and parameter values ​​will cause unsatisfactory detection results. Furthermore, the situation where the color difference changes greatly in the cigarette package area is a "forbidden zone" set by the traditional detection frame. In such an area, the color changes are large and the pattern is complex. It is difficult for traditional detection algorithms to identify defects in this area.

[0085] The main reason for equipment maintenance problems is the complex configuration of the detection algorithm, and the quality of personnel affects the effect of equipment maintenance. For the traditional cigarette package appearance inspection system, traditional machine vision manually selects the inspection area of ​​the cigarette package and manually configures it to achieve the purpose of inspection. For cigarette packages, it is necessary to set up many inspection areas on the surface of the cigarette package image, and the setting parameters of each inspection area are completely different, which greatly increases the complexity of the configuration parameters.

[0086] The cigarette package appearance quality inspection device of the disclosed embodiment inspects the overall appearance of the cigarette package without adding any additional inspection frame, and has strong maintainability.

[0087] In addition, the traditional cigarette package appearance inspection system only supports binary statistics of right and wrong, while the cigarette package appearance quality inspection device of the embodiment of the present disclosure can support multivariate classification statistics of defect types.

[0088] Traditional cigarette package appearance inspection systems are not easy to inspect areas with complex patterns or obvious color changes, but the cigarette package appearance quality inspection device of the disclosed embodiment can inspect such areas. For example, the traditional cigarette package appearance inspection system calculates the mean of the pattern or color. If defects appear in areas with obvious color changes or complex patterns, the mean change may not be obvious. The cigarette package appearance quality inspection device of the disclosed embodiment performs convolution and other operations on the pattern to extract the features of the pattern. The defect features that appear in complex areas are different from the features of the correct cigarette package, so it can detect areas with complex patterns or obvious color changes.

[0089] The conventional cigarette pack appearance inspection system can only reject all or none of the connected packs, whereas the cigarette pack appearance quality inspection device of the disclosed embodiment can reject the cigarette packs with appearance defects in the connected packs.

[0090] In addition, the cigarette package appearance quality inspection device of the embodiment of the present disclosure can support network deployment, and the inspection equipment can be networked throughout the factory, which has strong scalability. The traditional cigarette package appearance inspection system requires each device to be configured separately, while the cigarette package appearance quality inspection device of the embodiment of the present disclosure can support one model to be issued to all inspection devices for use. In addition, the hardware of the cigarette package appearance quality inspection device of the embodiment of the present disclosure is standardized, so that the equipment is easy to deploy and maintain quickly.

[0091] The artificial intelligence processing unit of the embodiment of the present disclosure can run on a Linux operating system and is not vulnerable to external attacks.

[0092] The principles of the image detection model of the embodiment of the present disclosure are described in detail below with reference to the accompanying drawings.

[0093] Figure 4 is a schematic diagram illustrating the principle of artificial intelligence deep learning according to some embodiments of the present disclosure.

[0094] The process of artificial intelligence deep learning can include: data input; basic feature extraction; multi-layer complex feature extraction; weight learning; and prediction results.

[0095] Based on an in-depth study of algorithm defects, the cigarette package appearance quality inspection device of the disclosed embodiment adopts deep learning technology to perform appearance quality inspection on cigarette packages. By having the machine learn features by itself, it eliminates human intervention in feature selection and improves detection accuracy.

[0096] One of the core problems that deep learning solves is to automatically combine simple features into more complex features and use these combined features to solve problems. In addition to learning the relationship between features and tasks, deep learning can also automatically extract more complex features from simple features. Deep learning can gradually transform simple features into more complex features layer by layer, making images of different categories more separable.

[0097] Figure 5 is a schematic diagram illustrating feature extraction by deep learning according to some embodiments of the present disclosure. Figure 5 It shows that deep learning algorithms can gradually combine more effective complex features such as lines, edges, corners, simple shapes and complex shapes from the pixel features of the collected cigarette package images.

[0098] Figure 6 is a schematic diagram showing the structure of a convolutional neural network according to some embodiments of the present disclosure.

[0099] The core technology of deep learning is the neural network structure. Figure 6 As shown, when distinguishing spiral data, the corresponding network is constructed. Figure 6 The left image data in is input into the network, the network is trained, the features are analyzed, and the classification results are output in the output layer on the right.

[0100] Preferably, when selecting a network, a convolutional neural network is used. A convolutional neural network can directly recognize visual patterns from the original image and is robust to geometric deformation. The advantages of a convolutional neural network are as follows: (1) the topological structure of the input image and the network can be well matched; (2) feature extraction and pattern classification are performed simultaneously; (3) weight sharing technology greatly reduces the training parameters of the network, making the structure of the neural network simpler and more adaptable. A convolutional neural network is a feedforward neural network that can extract its topological structure from a two-dimensional image, use a back propagation algorithm to optimize the network structure, and solve unknown parameters in the network. This network is often used to study two-dimensional image recognition problems. The network only requires a small amount of preprocessing, has a wide recognition range, and can tolerate image distortion. The classification model of a convolutional neural network is different from traditional image recognition in that it can directly input a two-dimensional image into the model, and then give the classification result at the output end. Its advantage is that it does not require complex preprocessing, and completely puts feature extraction and pattern classification into a black box. Through continuous optimization, the parameters required by the network are obtained, and the required classification is given at the output layer. The core of the network is the structural design of the network and the solution of the network.

[0101] Figure 7 is a schematic diagram showing a model of a convolutional neural network according to some embodiments of the present disclosure.

[0102] The structure of the convolutional neural network is as follows Figure 7 As shown in the figure, in addition to the input and output layers, the network also has intermediate convolutional layers, sampling layers and fully connected layers. The image is input from the input layer Input, passes through alternating sampling layers (for example, sampling layers S1 and S3) and convolutional layers (for example, convolutional layers C2 and C4) and the final fully connected layer (for example, fully connected layer F5), and the output of the network is given at the output layer.

[0103] Figure 8 It is a schematic diagram showing that data according to some embodiments of the present disclosure converges quickly after passing through the convolutional layer and the hidden layer.

[0104] The input layer normalizes the input data. The purpose of normalization is to make data of different dimensions converge quickly after passing through the convolutional layer and hidden layer. Figure 8 As shown. Taking image data as an example, the input data is pixels, and two-dimensional pixels can be divided into RGB channels. The standard deviation of each channel is calculated using a standardized method, such as the commonly used Z-Score standard score method, and then the pixels and standard deviation of each channel are normalized. The result data of the input layer will be output to the convolution layer.

[0105] There are several feature maps in the convolution layer and the sampling layer, and each feature map is connected to the feature map of the previous layer. In the convolution layer, the feature map of the previous layer is convolved with a learnable convolution kernel, and the output of the convolution result after the activation function forms the neurons of this layer, thus forming the feature map of this layer. Figure 7 As shown, the convolutional layers are C2 and C4, which appear alternately with the sampling layer. Each feature map output by the convolutional layer may establish a relationship with the convolution of several feature maps of the previous layer. In general, the calculation relationship of the convolutional layer is:

[0106]

[0107] Among them, l represents the layer, Kernel is the convolution kernel, each feature map can have a different convolution kernel, and M j is a selection of the input feature map, and each layer has a unique offset B. Therefore, the main task of the convolution layer is to select the angular features of the feature map of the previous layer from different angles to make it shift invariant. From the analysis of formula (1), we can see that the essence of convolution is to process the feature map of the previous layer to obtain the feature map of this layer. That is, if we want to obtain a neuron in the feature map of layer l, Need to use convolution kernel The neurons in the previous layer The nearby neurons perform convolution operations, such as Fig. 9 shown.

[0108] The sampling layer is relatively simple. Its main function is to reduce the spatial resolution of the network. By reducing the spatial resolution of the image, the offset and image distortion can be eliminated. The calculation formula of the neuron X on the sampling layer S is:

[0109]

[0110] Among them, n represents the window size from the convolution layer to the sampling layer. It is worth noting that S1 is actually a virtual layer and does not need to do too much operation. At this time, the sampling window is 1, that is, n = 1. The difference is that S3, whose sampling window is 2×2, then the feature map is sampled from C2 (28×28) to S3 (14×14). From the analysis of formula (2), we can see that the sampling process is as follows: Fig.10 As shown in the figure, the convolution layer and the sampling layer appear alternately, and each output feature map of the convolution layer may establish a relationship with the convolution of several feature maps of the previous layer. The main function of the sampling layer is to reduce the spatial resolution of the network, and to eliminate the offset and image distortion by reducing the spatial resolution of the image.

[0111] Convolutional neural networks will achieve invariance to displacement, scaling, and distortion of recognized images through three methods, namely local receptive field, weight sharing, and subsampling. The local receptive field means that the neurons in each layer of the network are only connected to the neural units in a small neighborhood of the previous layer. Through the local receptive field, each neuron can extract primary visual features, such as directional segments, endpoints, and corners. Weight sharing allows convolutional neural networks to have fewer parameters and require relatively less training data. Subsampling can reduce the resolution of features and achieve invariance to displacement, scaling, and other forms of distortion. A subsampling layer is usually used after the convolution layer to reduce computing time and establish spatial and structural invariance. This method is very simple and effective.

[0112] After the network is constructed, it is necessary to solve the network. If the parameters are assigned like the traditional neural network, each connection will have unknown parameters. The convolutional neural network uses weight sharing. In this way, the free parameters can be greatly reduced by sharing the same weights by the neurons on a feature map. This can be used to detect the effect of the same feature at different angles. From the above analysis, it can be seen that in network design, sampling layers and convolutional layers usually appear alternately. The previous layer of the fully connected layer is usually a convolutional layer. There will be a virtual S1 layer in the design to simply process the original image. At this time, no sampling operation is performed. Then this feature map will have 1×(1×1)×(32×32)×2 connections. When weight sharing is used, there are only 2 unknown parameters. In this way, the number of parameters will be 112309 if the traditional network connection method is used. If it is solved directly, the complexity of the network will be greatly increased. The number of unknown parameters of the convolutional neural network using weight sharing will be greatly reduced to 4327. The number of connections and unknown parameters of each layer are shown in Table 1.

[0113] Table 1 Connections and unknown parameters of convolutional neural networks

[0114]

[0115]

[0116] Local field techniques of convolutional neural networks such as Fig.11 It means that the neuron X in the convolution layer C needs to be connected to a local area of ​​the previous layer input Input to get the output at this time. Similarly, the neurons in the sampling layer S need to be connected to a local field of the previous convolution layer C to obtain the neurons of the current layer to form the feature map at this time. The neurons on each layer are the results obtained after the local field area of ​​the corresponding previous layer passes through the excitation function, and the results are given in the final output layer. The number of neurons in the output layer is the number of classifications.

[0117] In convolutional neural networks, weight updates are based on the back-propagation algorithm, in which the weights are updated from the weight w(t) at time t to the weight w(t+1) at time t+1, as shown below:

[0118] w(t+1)=w(t)+ηδ(t)x(t) (3)

[0119] Among them, η is the learning rate, x(t) is the input of the neuron, and δ(t) is the error term of the neuron. The error term has different expressions for output neurons and hidden neurons, and the error term can usually be given by gradient descent.

[0120] Convolutional neural networks do not require complex feature extraction when processing two-dimensional images. Therefore, cigarette package images can be directly input into the neural network, greatly reducing the difficulty of preprocessing; its local field and weight sharing technology reduces the parameter space and greatly reduces the complexity of the algorithm; subsampling technology enhances the robustness of the network and can tolerate a certain degree of image distortion.

[0121] In actual use, the various parameters of the convolutional neural network are trained by continuously providing sample sets.

[0122] Fig.12 is a flow chart showing a method for detecting the appearance quality of cigarette packages according to some embodiments of the present disclosure. Fig.12 As shown, the method includes steps S1202 to S1214.

[0123] In step S1202, when the cigarette pack is transported to the first position, the first positioning sensor detects the cigarette pack and sends information that the cigarette pack is transported to the first position to the controller.

[0124] In step S1204, after receiving the information that the cigarette pack is transported to the first location, the controller sends an image acquisition instruction to the image acquisition device.

[0125] In step S1206, the image acquisition device acquires image information of the cigarette package after receiving the image acquisition instruction, and sends the image information of the cigarette package to the artificial intelligence processing unit.

[0126] In step S1208, the artificial intelligence processing unit detects the image information of the cigarette package using the image detection model. When it is detected that there are defects in the appearance quality of the cigarette package, a rejection mark information is set for the cigarette package and the rejection mark information is sent to the controller.

[0127] In step S1210, when the cigarette pack is transported to the second position, the second positioning sensor detects the cigarette pack and sends information that the cigarette pack is transported to the second position to the controller, wherein the cigarette pack is transported from the first position to the second position by the conveyor belt.

[0128] In step S1212, after receiving the information that the cigarette pack is transported to the second position, the controller detects whether rejection mark information is set for the cigarette pack, and sends a rejection instruction to the rejection device if the rejection mark information is set for the cigarette pack.

[0129] In step S1214, the rejecting device performs a rejecting operation on the cigarette pack after receiving the rejecting instruction.

[0130] So far, a cigarette package appearance quality detection method according to some embodiments of the present disclosure is provided. The cigarette package appearance quality detection method detects the cigarette package appearance quality by using an image detection model, thereby improving the detection accuracy of the cigarette package appearance quality.

[0131] In some embodiments, the cigarette package appearance quality detection method may further include: training an image detection model using image information of cigarette packages with known appearance quality. The image detection model corresponds to the brand of the cigarette package.

[0132] In some embodiments, the step of training the image detection model includes: when there is no corresponding image detection model for the brand of the cigarette package, turning off the rejection device, establishing the brand of the cigarette package and loading the brand; adjusting the parameters of the image acquisition device so that the image acquisition device is in the best state for imaging the cigarette package; and using the image acquisition device to capture image information of the cigarette package and input it into the artificial intelligence processing unit to train the image detection model.

[0133] In other embodiments, the step of training the image detection model includes: when there is a corresponding image detection model for the brand of the cigarette package, using an image acquisition device to collect image information of the cigarette package and input it into the artificial intelligence processing unit to optimize the image detection model.

[0134] In the above embodiment, the new brand does not have an image detection model, so it is necessary to first shut down the rejection device, create a new brand name and load the brand, adjust the parameters of the image acquisition device to make the image imaging optimal, and then collect the training set, and manually judge the quality of the cigarette package image for training. The training of the existing detection model is to continuously optimize the detection model. The images collected by the equipment will be added to the materials in the model training, re-screened and added to the existing training set, and then the training operation will be performed.

[0135] In some embodiments, the above method may include a training program and an inference program, the training program collects pictures to train the model, and the inference program uses the trained model to detect real-time images.

[0136] In some embodiments, the cigarette package appearance quality detection method may further include: after receiving information that the cigarette package is transported to the first position, the controller sends a lighting instruction to the light source so that the light source illuminates the cigarette package. For example, when the cigarette package arrives at the first position, after receiving the cigarette package arrival signal sent by the first positioning sensor, the controller controls the light source to illuminate on the one hand, and sends an image acquisition instruction to the image acquisition device on the other hand. The image acquisition device uses three cameras to shoot the five sides of the cigarette package.

[0137] In some embodiments, the method for detecting the appearance quality of cigarette packs may further include: the auxiliary positioning sensor triggers a position signal when detecting a cigarette pack being transported to a corresponding position, and transmits the position signal to a controller; the controller obtains the number n of cigarette packs located before the cigarette pack with defective appearance quality based on the received position signal, and after receiving the information sent by the second positioning sensor that the n+1th cigarette pack is transported to the second position, the controller issues a rejection instruction when a rejection mark information is set for the cigarette pack, wherein n is an integer greater than or equal to 0.

[0138] In some embodiments, the rejected cigarette packs can be manually inspected visually, and combined with historical query records of defective images to determine whether the cigarette packs are falsely detected. If false detections are too high, the existing image detection model is iteratively trained until the results are satisfactory.

[0139] Fig.13 13 is a schematic diagram showing the structure of a cigarette pack appearance quality detection device according to other embodiments of the present disclosure. The cigarette pack appearance quality detection device includes a memory 1310 and a processor 1320.

[0140] in:

[0141] The memory 1310 may be a disk, a flash memory, or any other non-volatile storage medium. The memory is used to store Fig.12 The instructions in the corresponding embodiment.

[0142] The processor 1320 is coupled to the memory 1310 and can be implemented as one or more integrated circuits, such as a microprocessor or a microcontroller. The processor 1320 is used to execute instructions stored in the memory, and detect the appearance quality of the cigarette package using an image detection model, thereby improving the detection accuracy of the appearance quality of the cigarette package.

[0143] In some embodiments, the cigarette package appearance quality inspection device may include multiple memories 1310 and multiple processors 1320, and the multiple memories 1310 and the multiple processors 1320 may be matched and arranged in different components.

[0144] In some embodiments, it is also possible to Fig.14 As shown, the cigarette package appearance quality inspection device 1400 includes a memory 1410 and a processor 1420. The processor 1420 is coupled to the memory 1410 via a BUS 1430. The cigarette package appearance quality inspection device 1400 can also be connected to an external storage device 1450 via a storage interface 1440 to call external data, and can also be connected to a network or another computer system (not shown) via a network interface 1460, which will not be described in detail here.

[0145] In this embodiment, data instructions are stored in a memory, and the instructions are processed by a processor, and the appearance quality of the cigarette package is detected by an image detection model, thereby improving the detection accuracy of the appearance quality of the cigarette package.

[0146] In some embodiments, the cigarette package appearance quality detection device may include multiple memories 1410, multiple processors 1420, multiple buses 1430, multiple storage interfaces 1440, multiple external storage devices 1450 and multiple network interfaces 1460. The multiple memories 1410, multiple processors 1420, multiple buses 1430, multiple storage interfaces 1440, multiple external storage devices 1450 and multiple network interfaces 1460 can be matchedly arranged in different components.

[0147] In some embodiments, the present disclosure further provides a cigarette packing machine, which includes the cigarette pack appearance quality detection device as described above.

[0148] In some other embodiments, the present disclosure further provides a computer-readable storage medium having computer program instructions stored thereon, which are executed by a processor to implement Fig.12The steps of the method in the corresponding embodiment. Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, devices, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable non-transient storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0149] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0150] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0151] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0152] So far, the present disclosure has been described in detail. In order to avoid obscuring the concept of the present disclosure, some details known in the art are not described. Based on the above description, those skilled in the art can fully understand how to implement the technical solution disclosed here.

[0153] Although some specific embodiments of the present disclosure have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present disclosure. It should be understood by those skilled in the art that the above embodiments may be modified without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.

Claims

1. A cigarette package appearance quality inspection device, comprising: a first positioning sensor, configured to detect the cigarette pack when the cigarette pack is transported to the first position, and send information that the cigarette pack is transported to the first position to a controller; a second positioning sensor, for detecting the cigarette pack when the cigarette pack is transported to a second position, and sending information that the cigarette pack is transported to the second position to the controller, wherein the cigarette pack is transported from the first position to the second position by a conveyor belt; The controller is used to issue an image acquisition instruction after receiving the information that the cigarette package is transported to the first position, detect whether the cigarette package is set with rejection mark information after receiving the information that the cigarette package is transported to the second position, and issue a rejection instruction when the cigarette package is set with the rejection mark information; An image acquisition device, used for acquiring image information of the cigarette package after receiving the image acquisition instruction; an artificial intelligence processing unit, configured to receive the image information of the cigarette package from the image acquisition device, detect the image information of the cigarette package using an image detection model, and set rejection mark information for the cigarette package when it is detected that the appearance quality of the cigarette package is defective, and send the rejection mark information to the controller; a rejecting device, configured to perform a rejecting operation on the cigarette pack after receiving the rejecting instruction from the controller; and an auxiliary positioning sensor, arranged beside the conveyor belt and between the first positioning sensor and the second positioning sensor, for triggering a position signal when detecting a cigarette pack transported to a corresponding position, and transmitting the position signal to the controller; The controller is used to obtain the number n of cigarette packs located before the cigarette pack with defective appearance quality based on the received position signal, and after receiving the information sent by the second positioning sensor that the n+1th cigarette pack is transported to the second position, issue a rejection instruction when the cigarette pack is set with rejection mark information, wherein n is an integer greater than or equal to 0.

2. The cigarette package appearance quality inspection device according to claim 1, wherein: The artificial intelligence processing unit is also used to train the image detection model using image information of cigarette packages with known appearance quality.

3. The cigarette package appearance quality inspection device according to claim 1, wherein: The artificial intelligence processing unit is also used for not setting rejection mark information on the cigarette package when it is detected that there is no defect in the appearance quality of the cigarette package; The controller is also used for not issuing a rejection instruction when no rejection mark information is set for the cigarette package.

4. The cigarette package appearance quality inspection device according to claim 1, wherein: The image acquisition device comprises: a first camera, configured to take photos of a first side surface and a second side surface of the cigarette packet to obtain image information of the first side surface and the second side surface, and send the image information of the first side surface and the second side surface to the artificial intelligence processing unit, wherein the first side surface is adjacent to the second side surface; a second camera, used for taking photos of a third side surface and a fourth side surface of the cigarette package to obtain image information of the third side surface and the fourth side surface, and sending the image information of the third side surface and the fourth side surface to the artificial intelligence processing unit, wherein the third side surface is opposite to the first side surface, and the fourth side surface is opposite to the second side surface; and The third camera is used to take a picture of the top surface of the cigarette package to obtain image information of the top surface, and send the image information of the top surface to the artificial intelligence processing unit.

5. The cigarette package appearance quality inspection device according to claim 1, further comprising: a light source, for illuminating the cigarette pack after receiving an illumination instruction from the controller; Wherein, the controller is further used to issue the lighting instruction after receiving the information that the cigarette package is transported to the first position.

6. The cigarette package appearance quality inspection device according to claim 1, wherein: The first positioning sensor is a detection position optical fiber sensor, and the first position is a detection position; The second positioning sensor is a reject position optical fiber sensor, and the second position is a reject position; The controller includes an input and output control board; The artificial intelligence processing unit includes an industrial computer or a server.

7. The cigarette package appearance quality inspection device according to claim 1, further comprising: A display, used to display image information of the cigarette pack and statistical information of cigarette pack rejection; The artificial intelligence processing unit is further used to collect statistics on the situations of cigarette pack rejection to obtain the statistical information, and send the statistical information to the display.

8. The cigarette package appearance quality inspection device according to claim 7, further comprising: Switches, used to transmit data; Among them, the controller is communicatively connected to the image acquisition device through the switch, the image acquisition device is communicatively connected to the artificial intelligence processing unit and the display through the switch, and the artificial intelligence processing unit is communicatively connected to the controller and the display through the switch.

9. The cigarette package appearance quality inspection device according to claim 1, The image detection model corresponds to the brand of the cigarette pack; The image detection model is a convolutional neural network.

10. A method for detecting the appearance quality of a cigarette package, comprising: When the cigarette pack is transported to the first position, the first positioning sensor detects the cigarette pack and sends information that the cigarette pack is transported to the first position to the controller; The controller sends an image acquisition instruction to the image acquisition device after receiving the information that the cigarette package is transported to the first position; The image acquisition device acquires the image information of the cigarette package after receiving the image acquisition instruction, and sends the image information of the cigarette package to the artificial intelligence processing unit; The artificial intelligence processing unit detects the image information of the cigarette package using an image detection model, and when it is detected that the appearance quality of the cigarette package is defective, sets rejection mark information for the cigarette package, and sends the rejection mark information to the controller; When the cigarette pack is transported to the second position, the second positioning sensor detects the cigarette pack and sends information that the cigarette pack is transported to the second position to the controller, wherein the cigarette pack is transported from the first position to the second position by the conveyor belt; The controller detects whether the cigarette pack is set with rejection mark information after receiving the information that the cigarette pack is transported to the second position, and sends a rejection instruction to the rejection device if the cigarette pack is set with rejection mark information; and The rejection device performs a rejection operation on the cigarette pack after receiving the rejection instruction; Wherein, the cigarette package appearance quality detection method further includes: The auxiliary positioning sensor triggers a position signal when detecting a cigarette package transported to a corresponding position, and transmits the position signal to the controller; The controller obtains the number n of cigarette packs before the cigarette pack with defective appearance quality based on the received position signal, and after receiving the information sent by the second positioning sensor that the n+1th cigarette pack is transported to the second position, issues a rejection instruction when the cigarette pack is set with rejection mark information, wherein n is an integer greater than or equal to 0.

11. The cigarette package appearance quality detection method according to claim 10, further comprising: The image detection model is trained using image information of cigarette packages with known appearance quality, wherein the image detection model corresponds to the brand of the cigarette packages.

12. The cigarette package appearance quality inspection method according to claim 11, wherein: The steps of training the image detection model include: In the case where there is no corresponding image detection model for the brand of the cigarette package, shutting down the rejection device, establishing the brand of the cigarette package and loading the brand; adjusting the parameters of the image acquisition device so that the image acquisition device is in an optimal state for imaging the cigarette package; and The image acquisition device is used to acquire image information of cigarette packages and input the information into the artificial intelligence processing unit to train the image detection model.

13. The cigarette package appearance quality inspection method according to claim 11, wherein: The steps of training the image detection model include: In the case where there is a corresponding image detection model for the brand of cigarette pack, the image information of the cigarette pack is collected by the image acquisition device and input into the artificial intelligence processing unit to optimize the image detection model.

14. The cigarette package appearance quality detection method according to claim 10, further comprising: After receiving the information that the cigarette package is transported to the first position, the controller sends a lighting instruction to the light source so that the light source illuminates the cigarette package.

15. A cigarette package appearance quality inspection device, comprising: Memory; as well as A processor coupled to the memory, the processor being configured to execute the method according to any one of claims 10 to 14 based on instructions stored in the memory.

16. A cigarette packing machine, comprising: The cigarette pack appearance quality inspection device according to any one of claims 1 to 9 or the cigarette pack appearance quality inspection device according to claim 15.

17. A computer-readable storage medium having computer program instructions stored thereon, which implement the method according to any one of claims 10 to 14 when executed by a processor.

Images