Paperboard detection system based on machine vision

Abstract

The invention discloses a paperboard detection system based on machine vision. The paperboard detection system comprises a paperboard conveying system. The light source and the image acquisition device are matched for use to acquire a paperboard image; the analysis processing device is used for analyzing and identifying whether flaws exist in the paperboard image or not; when the analyzing and processing device recognizes that flaws exist in the paperboard images, the image recording system records and stores flaw pictures, meanwhile, the paperboard conveying system stops running to mark the flawed paperboards, and then the paperboard conveying system starts running again; the timer is used for recording the starting time T0 when the conveying platform runs for the first time, the moment Ti of pause every time and the moment Ti'of restarting running after pause every time; and the flaw position calculation device is used for calculating the position Li of the ith flaw on the paperboard according to the time recorded by the timer. Compared with the prior art, the scheme can automatically, quickly and accurately detect flaws on the surface of the paperboard, is high in detection efficiency and has an online repairing function.

Description

Technical Field

[0001] The invention belongs to the technical field of defect detection in a paperboard production process, and in particular relates to a paperboard detection system based on machine vision. Background Art

[0002] In the cardboard production process, the detection of defects on the cardboard surface is an important part of ensuring product quality. For cardboard products with defects, targeted treatment needs to be carried out according to the defect situation. For example, if there are stains or impurities on the cardboard surface, it is necessary to use a cleaning device to remove the stains, or cover minor stains through local coating; if there are bubbles on the cardboard surface, the bubble area needs to be compacted; if the moisture content of the cardboard is uneven, it needs to be humidified again; if the cardboard is warped or deformed, it needs to be flattened or the humidity control parameters need to be adjusted; if there is color difference in the local area of ​​the cardboard, it needs to be covered by local coating; if the coating on the cardboard surface is uneven, it needs to be coated again. However, there are many shortcomings in traditional detection methods. There are still problems such as low detection efficiency, high labor intensity of defect processing, and insufficient automation in the judgment and treatment of cardboard defects. The main reason is that the value of cardboard is low, the defect ratio is small compared to the overall, and there are many defect categories with strong randomness. Manual repair or discarding of defective cardboard is a more common method. However, with the application of functional cardboard, such as anti-static, antibacterial and other special coatings that meet special application requirements, the integration of QR code / RFID, the demand for high-performance cardboard such as waterproof, pressure-resistant and high-temperature resistant, new requirements have been put forward for multi-layer composite and automated control processes to achieve a more automatic, fast and accurate system for detecting cardboard surface defects. Summary of the invention

[0003] In order to solve the above problems, the present invention provides a cardboard detection system based on machine vision, which can automatically, quickly and accurately detect defects on the surface of the cardboard, has high detection efficiency, and can eliminate the influence of human factors on the detection results.

[0004] The technical solution adopted by the present invention is:

[0005] The present application discloses a cardboard detection system based on machine vision, comprising: a cardboard conveying system, which is used to convey cardboard and receive trigger information from an analysis and processing device to realize the trigger start and stop of cardboard conveying;

[0006] An array light source and an image acquisition device, wherein the array light source and the image acquisition device are used together to acquire a paperboard image;

[0007] An analysis and processing device is connected to the image acquisition device and the paperboard conveying system in communication; the analysis and processing device is used to analyze and identify defects in the paperboard image and feed back trigger information to the paperboard conveying system;

[0008] An image recording system is connected to the analysis and processing device in communication; when the analysis and processing device identifies defects in the cardboard image, the image recording system records and saves the defect image, and the cardboard conveying system pauses to mark or repair the defective cardboard. After the marking is completed or the repair is completed, the cardboard conveying system starts to run again; a timer is used to record the start time T0 of the conveying platform when it is first run, the time of each pause T i , and the time T′ when the operation restarts after each pause i ; wherein i is a natural number of 1, 2, ..., n, and n represents the number of times the conveying platform is paused, that is, a total of n defects are detected during the conveying of the paperboard; a defect position calculation device is used to calculate the position L of the i-th defect on the paperboard according to the time recorded by the timer i , where L i Calculated according to the following formula:

[0009] L i =v×{(T o -T0)-{(T′1-T1)+(T′2-T2)+…+(T′ i-1 -T i-1 )}}+C;

[0010] Wherein, v is the conveying speed of the cardboard, and C is the relative position correction value between the conveying distance of the cardboard and the conveying distance of the conveying platform.

[0011] As an optional technical solution, it also includes a high-pressure air system, a water replenishment system, a spraying system and a position adjustment system; the high-pressure air system, the water replenishment system and the spraying system receive trigger information from the analysis and processing device; wherein the high-pressure air system, the water replenishment system and the spraying system respectively correspond to high-pressure air nozzles, water mist nozzles and paint nozzles; the position adjustment system includes a carrier; the position adjustment system obtains the position information fed back by the defect position calculation device, and adjusts the carrier to the target position; the carrier is located above the cardboard conveying section, and is used to carry the high-pressure air nozzle, water mist nozzle and paint nozzle, so that after the high-pressure air system, the water replenishment system and the spraying system receive the trigger information, they perform the corresponding cardboard defect repair operation.

[0012] As an optional technical solution, the analysis and processing device includes at least one image processing server, which contains a machine vision cardboard surface defect model algorithm and a defect repair database. The machine vision cardboard surface defect model algorithm is used to analyze and identify the defects of the cardboard image acquired by the image acquisition device; the defect repair database contains preset repair types and repair parameters for repairing defects; the repair parameters include jet pressure size and jet duration, water mist spray size and spray duration, paint spray color and spray duration.

[0013] As an optional technical solution, the analysis and processing device includes a PID algorithm, which is applied to the parameter adjustment of at least one of the high-pressure air system, the water replenishment system and the spraying system; the PID algorithm includes the formula logic:

[0014]

[0015] Among them, K P is the proportional gain, K i is the integral gain, K d is the differential gain, u(t) is the repair parameter function, e(t).

[0016] As an optional technical solution, the array light source includes a plurality of LED light bars, each of which is distributed along the conveying direction of the cardboard, and the distance between two adjacent LED light bars is consistent.

[0017] As an optional technical solution, the image acquisition device includes multiple 4K ultra-high-definition cameras, the same number as the LED light strips, each of the 4K ultra-high-definition cameras is equipped with a C-mount large-target lens, and each of the 4K ultra-high-definition cameras is communicatively connected to the analysis and processing device to transmit the acquired cardboard image to the analysis and processing device for analysis and identification.

[0018] As an optional technical solution, the image recording system is communicatively connected to the analysis and processing device via a switch.

[0019] As an optional technical solution, the machine vision cardboard surface defect model algorithm includes two processes: image preprocessing and image secondary processing; wherein, the image preprocessing includes: grayscale processing and image enhancement processing of the collected cardboard image to enhance the image contrast and highlight the texture details of the scratches; then mean filtering the enhanced image to remove image noise interference; then dynamic threshold segmentation is performed on the mean filtered image. Since non-defect features are mostly block-shaped, and the detected defects are mostly long strips, the length and width features are used to extract the long strip defect area.

[0020] As an optional technical solution, the secondary image processing includes: performing image dilation processing on the defective area to further enhance the defect characteristics based on the defective area; performing mean filtering and dynamic threshold segmentation again to remove block details using the maximum circumscribed circle within a limited range; if long strips of scratches remain after the block details are removed, then outputting a defective detection result, otherwise, outputting a non-defective detection result.

[0021] The beneficial effects of the present invention are as follows: the cardboard detection system based on machine vision disclosed in the present application collects clear cardboard images through the cooperation of a light source and an image acquisition device, and then transmits the collected cardboard images to an analysis and processing device for analysis and identification. If the identification result is that there is a defect, the image recording system records and saves the defect image, and the cardboard conveying system is suspended to mark or repair the defective cardboard; after the marking or repair is completed, the cardboard conveying system runs again. It realizes the rapid and accurate detection of defects on the cardboard surface, improves the detection efficiency of defects, and realizes convenient online repair through the coordination of the cardboard conveying system, high-pressure air system, water replenishment system and spraying system. Based on the high-pressure air system, combined with the high-pressure air nozzle, it can not only remove some impurities or dirt, but also pressurize and level the locally bubbled cardboard. Based on the water replenishment system, it can achieve local humidity adjustment, which is also helpful for auxiliary treatment of leveling and bubbling; based on the spraying system and different spraying materials, it is helpful to realize the realization of printing repair and functional layer repair; combined with the feedback of machine vision, it is helpful to quickly identify the type of defects, and record and analyze and make corresponding improvements in combination with production setting parameters, pre-finished product conditions, and repair conditions; based on the feedback information and algorithms, it is helpful to overcome the transient difference between machine vision and the target effect at the time of information acquisition due to differences in humidity or dryness, surface printing material conditions, angle conditions, etc., and it is not easy to perform quick repairs. Therefore, the application scheme improves production efficiency, reduces the labor intensity of inspection personnel, and eliminates the influence of human factors on the inspection results to the greatest extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a mechanical structure diagram of a cardboard detection system based on machine vision in an exemplary embodiment.

[0023] Figure 2 is an architectural diagram of a cardboard detection system based on machine vision in an exemplary embodiment.

[0024] Figure 3 The present invention is a flow chart of performing defect analysis and identification on a collected cardboard image in an exemplary embodiment. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. The components and method steps of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0026] Example

[0027] like Figure 1 and Figure 2 As shown, the present application discloses a cardboard detection system based on machine vision, comprising:

[0028] A cardboard conveying system is used to convey cardboard 1; wherein, the cardboard conveying system comprises an active rod 9, a driven rod 10 and a conveying controller, and the two ends of the cardboard are respectively located on the active rod 9 and the driven rod 10, and the active rod 9 rotates to drive the cardboard 1 to move. Specifically, the active rod 9 can be driven to rotate by a servo motor, and the servo motor is controlled by the conveying controller. The cardboard conveying system is provided with a position sensor to obtain the position information of the conveyed cardboard; the servo feedback signal of the servo motor and the signal of the position sensor are fed back to the conveying controller, and the conveying controller realizes the precise triggering start and stop of the cardboard conveying in combination with the information of the receiving analysis and processing device.

[0029] The array light source 2 and the image acquisition device are used together to acquire the cardboard image. The array light source 2 is set to provide a suitable light environment for the image acquisition device, which can display the defects on the cardboard more clearly. For example, when there is a scratch defect somewhere on the cardboard, the array light source 2 can be set to make the defect more obvious in the acquired cardboard image, so as to facilitate the subsequent analysis and processing device to analyze and identify the defects of the cardboard image. The array light source 2 includes a plurality of LED light bars, and the image acquisition device includes a plurality of 4K ultra-high-definition cameras 4.

[0030] The analysis and processing device 5 is connected to the image acquisition device and the paperboard conveying system in communication; the analysis and processing device is used to analyze and identify defects in the paperboard image and feed back trigger information to the paperboard conveying system.

[0031] The image recording system 7 is in communication with the analysis and processing device 5; when the analysis and processing device 5 identifies defects in the cardboard image, the cardboard conveying system is suspended, and the image recording system 7 records and saves the defective image. After the defective cardboard is marked, the cardboard conveying system starts to operate again. The marking of the defective cardboard can be done manually or by a separately set spraying system.

[0032] The display screen 8 is connected to the image recording system 7 for displaying the defect image transmitted by the image recording system 7 .

[0033] The timer is used to record the start time T0 of the first operation of the transmission platform, the time of each pause T i , and the time T′ when the operation restarts after each pause i ; wherein i is a natural number of 1, 2, ..., n, and n represents the number of times the conveying platform is paused, that is, a total of n defects are detected in the process of conveying the cardboard; specifically, the timer is connected to the servo motor for communication. When the cardboard is installed on the active rod 9 and the driven rod 10, the servo motor starts to drive the active rod 9 to rotate and then drive the cardboard to move. At this time, the timer records the start time T0 of the servo motor; each time a defect is detected, the conveying platform needs to be paused and the servo motor stops working. At this time, the timer records the time when the servo motor stops working T i When the defect is marked, the servo motor starts working again, and the timer records the time when the servo motor starts working again T′ i .

[0034] A defect position calculation device is used to calculate the position L of the i-th defect on the paperboard according to the time recorded by the timer. i , where L i Calculated according to the following formula:

[0035] L i =v×{(T i -T0)-{(T′1-T1)+(T′2-T2)+…+(T′ i-1 -T i -1)}}+C…(1)

[0036] Wherein, v is the conveying speed of the paperboard. In this embodiment, the servo motor rotates at a constant speed, and the paperboard moves at a constant speed. The defect position calculation device can be an independent controller, or can be implemented by an analysis and processing device.

[0037] Specifically, the defect position calculation device is connected to the timer in communication. After receiving the data recorded by the timer, the defect position calculation device starts to calculate the position L of the defect on the paperboard. iAlthough the defects are marked, some defects may still be missed in the subsequent defect processing according to the traditional method; and in the case of a long cardboard, even if the defects are marked, it will be very troublesome to determine the location of the defects only based on the marks when processing the defects. Therefore, in this embodiment, the location of the defects is determined by the above formula (1).

[0038] Take the case where three defects were detected in total as an example. The servo motor starts working at T0, and defects are detected at T1, T2, and T3. It pauses for (T′1-T1)s after T1, (T′2-T2)s after T2, and (T′3-T3)s after T3. The position of the defect on the cardboard is equal to the total distance the cardboard moves before the defect is detected. The position of the defect detected for the first time on the cardboard at T1 is:

[0039] L1=v×(T1-T0)

[0040] The position of the second defect detected at time T2 on the cardboard is:

[0041] L2 = v × {(T2-T0)-(T′1-T1))

[0042] The position of the defect detected for the third time at time T3 on the cardboard is:

[0043] L3=v×{(T3-T0)-{(T′1-T1)+(T′2-T2)}}

[0044] When dealing with defects later, the location of the defects on the cardboard can be determined based on the calculation results of L1, L2, and L3, which is very convenient and can avoid missing defects.

[0045] Here, the default value of C is 0. Generally speaking, when there is no cutting during the cardboard transmission, the relative correction value of the cardboard transmission distance and the transmission distance of the transmission platform is 0. When cutting occurs, it can be corrected manually or by referring to the size of the break interval fed back by the position sensor.

[0046] In this embodiment, the array light source 2 includes 3-5 LED light bars, such as Figure 1 As shown, in this embodiment, four LED light bars are used, each of which is distributed along the conveying direction of the cardboard 1, and the spacing between two adjacent LED light bars is consistent. The image acquisition device includes a plurality of 4K ultra-high-definition cameras 4, the number of which is the same as that of the LED light bars, each of which is equipped with a C-port large target lens 3, and each of which is connected to the analysis and processing device 5 for communication, so as to transmit the collected cardboard image to the analysis and processing device 5 for analysis and identification. Specifically, as Figure 1 As shown, in this embodiment, four 4K ultra high-definition cameras 4 are provided, and the analysis and processing device 5 includes an image processing server, and the four 4K ultra high-definition cameras 4 are communicatively connected to the image processing server one by one.

[0047] The image acquisition device further includes a support crossbar 11 and a longitudinal support bar 12, wherein the longitudinal support bar 12 is fixed to the support crossbar 11 and is located at the lower side of the paperboard 1; the plurality of 4K ultra-high-definition cameras 4 are evenly distributed on the longitudinal support bar 12. The image recording system 7 is communicatively connected to the analysis and processing device 5 via a switch 6.

[0048] As an optional implementation, it also includes a high-pressure air system, a water supply system, a spraying system and a position adjustment system; the high-pressure air system, the water supply system and the spraying system receive trigger information of the analysis and processing device; wherein the high-pressure air system, the water supply system and the spraying system respectively include a high-pressure air nozzle, a water mist nozzle and a paint nozzle;

[0049] The position adjustment system includes a carrier; the position adjustment system is generally realized by a position controller controlling a servo motor to drive an XY platform with a lead screw, and the carrier moves with the rotation of the lead screw. This part is the prior art; since the position controller of the position adjustment system obtains the position information fed back by the defect position calculation device and adjusts the carrier to the target position; through the bracket design of the XY platform, the carrier is located above the cardboard conveying section and carries the high-pressure air nozzle, water mist nozzle and paint nozzle. After the high-pressure air system, the water supply system and the spraying system receive the trigger information, the corresponding cardboard defect repair operation is performed.

[0050] As an optional implementation, the analysis and processing device 5 includes at least one image processing server. Figure 1 As shown, in this embodiment, the image processing server can be one or more. The setting of more than one server is helpful to improve the parallel speed and is suitable for high-speed and large-size cardboard detection. In this embodiment, there are four image processing servers, and each image processing server is respectively connected to a 4K ultra-high-definition camera 4. Each of the image processing servers contains a machine vision cardboard surface defect model algorithm and a defect repair database. The machine vision cardboard surface defect model algorithm is used to analyze and identify the defects of the cardboard image collected by the image acquisition device; the defect repair database contains the preset repair type and repair parameters for repairing defects; the repair parameters include the jet pressure size and jet duration, the water mist spray size and spray duration, and the paint spray color and spray duration.

[0051] As an optional implementation, the analysis and processing device includes a PID algorithm, and the PID algorithm is applied to the parameter adjustment of at least one of the high-pressure air system, the water replenishment system and the spraying system; the PID algorithm includes the formula logic:

[0052]

[0053] Among them, K P is the proportional gain, K i is the integral gain, K d is the differential gain, u(t) is the repair parameter function, and e(t) is the difference between the target repair effect and the actual repair effect.

[0054] Taking stains or blistering defects as an example, when the defective area is calculated through image analysis, the output result is only 32% of the clean area area (or the inverse logic of the stain color block area calculation) and there is a thickness shadow, it is considered that the high-pressure gas can handle the defect. The target value of the clean area area is above 95%, and the initial error size is 63%. The corresponding repair parameter function is calculated to convert the 63% cleanliness error value into the single-point position pressure output size of the high-pressure gas and the purge time of the clean area. When the pressure of the high-pressure gas is fixed, it simply corresponds to the time relationship. Therefore, the 63% initial cleanliness error value is converted into the error value e(t) between the target repair effect containing the maintenance parameter (jet time) and the actual repair effect. As the repair time goes by, the cleanliness error e(t) will gradually decrease until it is qualified; similar to the repair algorithm logic of the high-pressure air system, the water replenishment system and the spraying system are also converted into the difference between the spray time and the spraying time. The difference is that in order to achieve a better cleaning effect, the high-pressure air system can have multiple high-pressure air nozzles (such as one vertically downward and one tilted downward in the X / Y direction). It should be noted that when adjusting the position, the defect position coordinates need to be used as a reference, plus the position offset parameters of the nozzle (based on the relative positions of each nozzle on the stage).

[0055] Compared with the preset fixed database parameters, the PID algorithm helps to combine existing information with new detection and repair information, and continuously optimize the accuracy of repair parameters. When the detection statistics after the corresponding category of the defect pattern and the corresponding repair parameters are executed reach the qualified threshold, the formula application statistically qualified values ​​can be stored in the defect repair database. In this way, on the one hand, the roughly set parameters can be optimized, and on the other hand, the cardboard type detection can be expanded and more defect repair information can be enriched; on the other hand, the calculation requirements can be continuously simplified, especially the defect repair solution analysis logic can be simplified to a table lookup logic, further improving the running speed: when the analysis finds that the defect parameters exist in the defect database, there is no need to jump to run the PID algorithm, and the repair parameters in the defect database can be directly called and transmitted to the controller of the execution subject accordingly.

[0056] As an optional implementation, the analysis and processing device includes a reinforcement learning algorithm, which is applied to parameter adjustment of at least one of the high-pressure air system, the water replenishment system and the spraying system; the reinforcement learning algorithm includes the formula logic:

[0057] R(t)=-α·|r(t)-y(t)|-β·|w(t)-w(t-1)|;

[0058] Among them, R(t) is the reward function, r(t) represents the target repair effect, y(t) represents the actual repair effect, w(t) represents the current repair parameter, and α and β are weight coefficients used to balance the repair effect and parameter adjustment range.

[0059] Specifically, a deep Q network is used, and the input layer is state features, such as stain cleanliness, color difference value, coating uniformity, humidity, etc. For example, through image recognition analysis, the color difference value ΔE, coating uniformity U and cardboard humidity w are obtained.

[0060] Construct the state vector s t =[ΔE,U,w];

[0061] H represents the spray water replenishment amount (ml / s), P represents the paint spraying amount (g / s),

[0062] Construct the repair action vector a t =[H, P];

[0063] The reward function is:

[0064] R(t)=-α·|ΔE target -ΔE|-γ·|U target -U|-β·|w(t)-w(t-1)|

[0065] Among them, α, β, and γ are weight coefficients used to balance the priorities of different goals;

[0066] The hidden layer can use a convolutional neural network or a multi-layer fully connected network to extract features;

[0067] Each repair action a in the output layer t = Q value of [H, P] (expected cumulative reward);

[0068] Store historical data(s t , a t , r t ,s t+1…) in the experience pool, and update the network parameters by minimizing the mean square error of the Q value. Compared with the PID algorithm, the reinforcement learning algorithm will find the optimal combination faster in the collaborative repair of multi-parameter coordination, is suitable for multi-objective optimization, and has better self-learning ability.

[0069] As an optional implementation, Figure 3 As shown, the machine vision cardboard surface defect model algorithm includes two processes: image preprocessing and image secondary processing; wherein the image preprocessing includes:

[0070] Grayscale and image enhancement processing is performed on the collected cardboard images to enhance the image contrast and highlight the texture details of the scratches;

[0071] Then, the enhanced image is subjected to mean filtering to remove image noise interference;

[0072] Then, dynamic threshold segmentation is performed on the mean-filtered image. Since non-defect features are mostly block-shaped, and the detected defects are mostly long strips, the length and width features are used to extract the long strip defect area.

[0073] As an optional implementation, since the defective area still contains non-defective features similar to the defect, secondary processing is required, such as Figure 3 As shown, the secondary image processing includes:

[0074] Perform image dilation processing on the defect area to further enhance the defect features based on the defect area;

[0075] Perform mean filtering and dynamic threshold segmentation again, and use the maximum circumscribed circle within a limited range to remove blocky details;

[0076] If long strips of scratches remain after the block details are removed, a defective detection result is output; otherwise, a non-defective detection result is output; the algorithm process has a good scratch fitting effect and can identify the scratch area in the image.

[0077] As an embodiment, the linear speed of the cardboard 1 during operation can be set to 110-130 meters / minute, and in this embodiment, it is set to 120 meters / minute. The 4K ultra-high-definition camera 4 is equipped with a C-mouth large target surface lens 3, and the C-mouth large target surface lens 3 adopts a fixed-focus lens. The analysis and processing device 5 is embedded with a machine vision cardboard surface defect model algorithm, which uses the algorithm to analyze and identify the pictures uploaded by the 4K ultra-high-definition camera 4 to determine whether there are defects on the cardboard surface. When a defect is detected, the cardboard conveying system is suspended (that is, the servo electrode controls the active roller 9 to stop rotating), and the image recording system 7 will record the defective picture and display it on the display screen 8. When the online repair system is not set, the defective part can be marked on the defective cardboard, and the cardboard conveying system is restarted after the marking is completed (that is, the servo electrode controls the active roller 9 to start rotating again).

[0078] The 4K ultra-high-definition camera 4 has the characteristics of high resolution, rich color measurement and high color depth, high real-time measurement and high speed, and is suitable for high-speed large-format cardboard industrial inspection, and can replace traditional manual inspection methods. It can accurately capture images on the high-speed moving cardboard surface and has high-precision detail capture capabilities. The specific working process of the system in this embodiment is:

[0079] 1. The array light source 2 and the 4K ultra-high-definition camera 4 are turned on, and the cardboard conveying system is started. The cardboard 1 moves in the following direction: Figure 1 As shown by the middle arrow, the linear speed of the cardboard movement is 120 m / min.

[0080] 2. The 4K ultra-high-definition camera 4 captures the cardboard image in real time and sends it to the analysis and processing device 5. The analysis and processing device 5 processes the collected cardboard image to detect whether there is a defect. If the analysis and processing device 5 detects a defect, the cardboard conveying system will suspend operation, and the image recording system 7 will record the defect image and display it on the display screen 8. At the same time, the defective part is marked on the cardboard by a manual or automated robot arm, and the cardboard conveying system is restarted after the marking is completed.

[0081] The above description is only a specific implementation mode of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. All technical solutions that fall within the scope defined by the claims of the present invention fall within the protection scope of the present invention.

Claims

1. A cardboard detection system based on machine vision, characterized in that: include: The cardboard conveying system is used to convey the cardboard and receive the trigger information of the analysis and processing device to realize the trigger start and stop of the cardboard conveying; An array light source and an image acquisition device, wherein the array light source and the image acquisition device are used together to acquire a paperboard image; An analysis and processing device is connected to the image acquisition device and the paperboard conveying system in communication; the analysis and processing device is used to analyze and identify defects in the paperboard image and feed back trigger information to the paperboard conveying system; An image recording system is connected to the analysis and processing device for communication; when the analysis and processing device identifies that there are defects in the cardboard image, the image recording system records and saves the defective image, and the cardboard conveying system is suspended to mark or repair the defective cardboard, and the cardboard conveying system starts to operate again after the marking or repair is completed; The timer is used to record the start time T0 of the first operation of the transmission platform, the time of each pause T i , and the time T′ when the operation restarts after each pause i , where i is a natural number; A defect position calculation device is used to calculate the position L of the i-th defect on the paperboard according to the time recorded by the timer. i , where L i Calculated according to the following formula: L i =v×{(T i -T0)-{(T′1-T1)+(T′2-T2)+…+(T′ i-1 -T i-1 )}}+C; Wherein, v is the conveying speed of the cardboard, and C is the relative position correction value between the conveying distance of the cardboard and the conveying distance of the conveying platform.

2. The cardboard detection system based on machine vision according to claim 1, characterized in that: It also includes a high-pressure air system, a water supply system, a spraying system and a position adjustment system; the high-pressure air system, the water supply system and the spraying system receive trigger information of the analysis and processing device; wherein the high-pressure air system, the water supply system and the spraying system respectively include a high-pressure air nozzle, a water mist nozzle and a paint nozzle; The position adjustment system includes a carrier; the position adjustment system obtains the position information fed back by the defect position calculation device and adjusts the carrier to the target position; the carrier is located above the cardboard conveying section, carrying the high-pressure air nozzle, the water mist nozzle and the paint nozzle, so that after the high-pressure air system, the water replenishment system and the spraying system receive the trigger information, the cardboard defect repair operation is performed accordingly.

3. The cardboard detection system based on machine vision according to claim 2, characterized in that: The analysis and processing device includes at least one image processing server, which includes a machine vision cardboard surface defect model algorithm and a defect repair database. The machine vision cardboard surface defect model algorithm is used to analyze and identify the defects of the cardboard image collected by the image acquisition device; the defect repair database includes preset repair types and repair parameters for repairing defects; the repair parameters include jet pressure size and jet duration, water mist spray size and spray duration, paint spray color and spray duration.

4. The cardboard detection system based on machine vision according to claim 3, characterized in that: The analysis and processing device includes a PID algorithm, which is applied to the parameter adjustment of at least one system among the high-pressure air system, the water replenishment system and the spraying system; the PID algorithm includes the formula logic: Among them, K P is the proportional gain, K i is the integral gain, K d is the differential gain, u(t) is the repair parameter function, and e(t) is the difference between the target repair effect and the actual repair effect.

5. A cardboard detection system based on machine vision according to claim 3 or 4, characterized in that: The analysis and processing device includes a reinforcement learning algorithm, which is applied to the parameter adjustment of at least one of the high-pressure air system, the water replenishment system and the spraying system; the reinforcement learning algorithm includes the formula logic: R(t)=-α·|r(t)-y(t)|-β·||w(t)-w(t-1)||; Among them, R(t) is the reward function, r(t) represents the target repair effect, y(t) represents the actual repair effect, w(t) represents the current repair parameter, and α and β are weight coefficients used to balance the repair effect and parameter adjustment range.

6. A cardboard detection system based on machine vision according to claim 1 or 2, characterized in that: The array light source includes a plurality of LED light bars, each of which is distributed along the conveying direction of the paperboard, and the distance between two adjacent LED light bars is consistent.

7. The cardboard detection system based on machine vision according to claim 6, characterized in that: The image acquisition device includes a plurality of 4K ultra-high-definition cameras, the number of which is the same as that of the LED light strips. Each of the 4K ultra-high-definition cameras is equipped with a C-mount large-target lens, and each of the 4K ultra-high-definition cameras is communicatively connected to the analysis and processing device to transmit the acquired cardboard image to the analysis and processing device for analysis and identification.

8. A cardboard detection system based on machine vision according to claim 1 or 2, characterized in that: The image recording system is communicatively connected to the analysis and processing device via a switch.

9. The cardboard detection system based on machine vision according to claim 2, characterized in that: The machine vision cardboard surface defect model algorithm includes two processes: image preprocessing and image secondary processing; wherein the image preprocessing includes: Grayscale and image enhancement processing is performed on the collected cardboard images to enhance the image contrast and highlight the texture details of the scratches; Then, the enhanced image is subjected to mean filtering to remove image noise interference; Then, dynamic threshold segmentation is performed on the mean-filtered image. Since non-defect features are mostly block-shaped, and the detected defects are mostly long strips, the length and width features are used to extract the long strip defect area.

10. The cardboard detection system based on machine vision according to claim 9, characterized in that: The secondary image processing comprises: Perform image dilation processing on the defect area to further enhance the defect features based on the defect area; Perform mean filtering and dynamic threshold segmentation again, and use the maximum circumscribed circle within a limited range to remove blocky details; If long strips of scratches remain after the block details are removed, a defective test result is output; otherwise, a non-defective test result is output.

Images