A wrinkle detection method for valve bag packaging machine

Abstract

The present invention discloses a wrinkle detection method for valve bag packaging machines, including: adjusting the backlight module so that the image texture of the valve bag product is clearly displayed on the display, starting the linear array industrial camera module to collect the image of the valve bag product in real time to determine whether the image clarity meets the initial setting requirements; performing image segmentation on the image in the target frame and quantitatively evaluating the wrinkle information based on the area estimation of the pixel points; removing unqualified products on the conveyor line; recording data; inputting LSTM for network learning and assisting in adjusting the production parameters of the packaging machine according to the output feedback information. The present invention effectively provides an online detection solution to quickly remove unqualified products; adopts the technical means of segmenting the target image to achieve quantitative evaluation of wrinkle information, thereby improving the sorting automation level of wrinkle detection, reducing the labor intensity of workers, and ensuring the accuracy of wrinkle detection.

Description

Technical Field

[0001] The invention belongs to the technical field of image detection, and in particular relates to a wrinkle detection method for a valve bag packaging machine. Background Art

[0002] At present, competition among various products is very fierce. In order to improve product competitiveness, more and more products pay more attention to the aesthetics of product packaging. Once the valve bag packaging machine is sealing and packaging, due to uneven heating, wrinkles often appear at the sealing point. The main reason is that the temperature of the heat shrink machine is not adjusted properly, resulting in wrinkles. Wrinkles will not only affect the appearance of the packaged product, but also cause poor sealing at the sealing position. During transportation, it is easy to cause material spillage, leakage, etc., causing safety accidents during transportation.

[0003] Machine vision simulates the human eye to perform product inspection. With its economy and convenience, it has become the focus of research in engineering inspection. It can well replace the human eye to inspect products, reduce the mental fatigue of the human eye, effectively inspect the surface quality of packaging bags, and better realize the production model of integrated production and inspection. Summary of the invention

[0004] The present invention provides a wrinkle detection method for a valve bag packaging machine, which solves the above problems.

[0005] In order to solve the above technical problems, the present invention is achieved through the following technical solutions:

[0006] A wrinkle detection method for a valve bag packaging machine of the present invention is implemented by a wrinkle detection device, which is installed on the valve bag packaging machine and includes an industrial camera module, a backlight module and a telescopic cylinder. The industrial camera module is installed on the upper left of the valve bag packaging machine, the backlight module is installed on the right side of the valve bag packaging machine, the backlight module is vertically downward, and the telescopic cylinder is installed on one side of the valve bag discharge port. The wrinkle detection method includes the following steps:

[0007] Step 1: During the operation of the valve bag packaging machine, adjust the backlight module so that the image texture of the valve bag product is clearly displayed on the monitor;

[0008] Step 2: Start the linear array industrial camera module to collect the valve bag product image in real time and determine whether the image clarity meets the initial setting requirements;

[0009] Step 3: If the image clarity does not meet the initial setting requirements, the acquired image is enhanced and the enhanced image is output;

[0010] Step 4: If the image clarity meets the initial setting requirements, the collected image is subjected to YOLO forward inference to output the image wrinkle position information;

[0011] Step 5: Use the region segmentation and merging method to segment the image in the target frame, estimate the area based on the pixels, and quantitatively evaluate the wrinkle information;

[0012] Step 6: Establish the Kalman filter equation, track and detect products, and remove unqualified products on the conveyor line;

[0013] Step 7: Using the timestamp as the basic unit, record the work product data in real time and mark the work data where wrinkles appear on the seal;

[0014] Step 8: Input LSTM for network learning, and perform auxiliary adjustment on the production parameters of the packaging machine based on the output feedback information.

[0015] Furthermore, the image clarity specifically uses the average gradient as an image evaluation index. The average gradient can reflect the detail contrast and texture transformation in the image. The specific calculation is as follows:

[0016]

[0017] In the formula, ΔxF(i,j) and ΔyF(i,j) represent the first-order difference of pixel point (i,j) in the x or y direction respectively. By calculating a single product picture of a finished packaging bag, the clarity of the collected picture is evaluated to ensure that the texture features of the input image are clear.

[0018] Furthermore, the YOLO forward reasoning in the step 4 includes the training of the YOLOX-S detection model for valve pocket wrinkles. By manually collecting some images and collecting an open source data set, namely, a valve pocket aluminum sheet surface defect detection data set, 420 images are obtained. The wrinkle data set is cyclically generated using a style transfer method to balance the number of positive and negative samples. The existing sample data set is then expanded through geometric rotation, local splicing, and mirror flipping. The images are annotated using LabelImage software, and the annotated categories are divided into three types of targets: normal products, surface wrinkles, and sealing wrinkles. A wrinkle data set is established, and the data set is divided into a training set and a test set in a ratio of 7:3.

[0019] Furthermore, the YOLO forward reasoning of the image in the step 4 specifically adopts the YOLOX network structure of the improved YOLOX-S detection model, mainly including: removing the 13×13 feature map outputted from the three branches in the YOLO-Head part of the YOLOX-S detection model, which is responsible for the feature map of large targets, and introducing an attention mechanism module, and replacing the ordinary convolution operation in the basic module with a deep separable convolution operation.

[0020] Furthermore, the image is subjected to YOLO forward reasoning by inputting the data set into the YOLOX-S detection model in batches, inputting images with a width and height of 416×416×3, outputting feature layers of 26×26×8 and 52×52×8, and outputting feature layers with a length and width of 20, as well as the coordinate information (x, y, w, h) of the target boxes of three categories for prediction, and obtaining 3380×8 prediction box information through the Reshape operation. Through back propagation, the losses are calculated respectively, which mainly include three parts: bounding box loss function, Iou perception classification loss and category loss. Since the wrinkles of the packaging are too small to be detected in the collected image, the real box and the predicted box are often in an inclusive relationship. Therefore, the CIOU loss function is used as the bounding box loss function, and the CIOU loss function is used as the bounding box loss function; the original loss function is replaced, and the specific formula of the CIOU loss function is as follows:

[0021]

[0022]

[0023]

[0024] Among them, IoU is the intersection over union ratio of the predicted box and the real box, ρ represents the Euclidean distance between the two points, b and b gt They represent the center points of the predicted box and the real box respectively, c represents the diagonal distance between the predicted box and the minimum bounding rectangle of the real box, α is used as the coordination ratio parameter, γ is used to measure the length ratio consistency parameter of the box, and w gt , w and h gt , h represent the width and height of the real box and the predicted box respectively.

[0025] Furthermore, by setting the sparsity factor γ, the trained model is weighted and sorted, the maximum and minimum values ​​of the weights are removed, and the number of some convolution kernels in the middle convolution layer of the model is reduced through unstructured sparsity. The detection model is sparse and fine-tuned on the training dataset to restore the accuracy of the model. Finally, the YOLOX-S detection model is deployed in embedded terminal devices.

[0026] Furthermore, in the step five, the image is divided into a set of arbitrary non-overlapping initial areas, and these areas are split and merged according to a given uniformity detection criterion until the image is finally divided into the smallest number of uniform areas. After the image position of the product with wrinkles is determined, it is necessary to estimate the actual size of the wrinkles through pixel points. At the same time, by calculating the area of ​​the pixel points, the degree of wrinkles can be roughly divided into three types: severe, moderate, and slight.

[0027]

[0028] Among them, S g is the pixel area of ​​the rectangular frame detected by YOLOX, and S is the pixel area after segmentation.

[0029] Furthermore, in step 6, the Kalman filter equation is established for target tracking with wrinkle problem to complete the update of target position. Since there is no additional control input during the observation process, uk is zero. Since the system has no additional control input, the state equation is:

[0030] x k =Ax k-1 +w k-1

[0031] x k =(x,y,v x ,v y ) T

[0032]

[0033] The coordinates of the target wrinkle are (x, y). Assume that the position of the detected object in the image changes in a uniform linear motion, and the change speeds are v and x , v y , A is the state transfer matrix, w k-1 is the process noise; the observation equation is:

[0034] z k =Hx k +v k

[0035]

[0036] z k is the observed value measured by YOLOX, the coordinates of the center point of the target in the image, H is the observation matrix, v k Represents the actual noise in the measurement process, and completes the update of the target position through prediction.

[0037] Furthermore, in the step eight, the LSTM neural network is used to adjust the parameters of the equipment, mainly by inputting the working current of the packaging machine, the temperature at the sealing point, as well as the production rhythm of the product, the feeding rate, the actual quality of the packaged product and other actual control parameters, so as to assist in adjusting the various working parameters of the coating machine.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. The present invention inputs the collected images into the deep learning detection model based on YOLO, and segments and identifies the identified foreign body images. The lightweight detection model effectively provides an online detection solution to quickly eliminate unqualified products.

[0040] 2. The present invention adopts the technical means of segmenting the target image to achieve quantitative evaluation of wrinkle information, thereby improving the level of sorting automation of wrinkle detection, reducing the labor intensity of workers, and ensuring the accuracy of wrinkle detection.

[0041] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0043] Figure 1 Schematic diagram of the process of the method corresponding to the specific embodiment 1 of the present invention;

[0044] Figure 2 A schematic diagram of the process of the wrinkle detection method of the present invention;

[0045] Figure 3 It is a structural schematic diagram of a wrinkle detection device corresponding to the wrinkle detection method of the present invention;

[0046] In the accompanying drawings, the components represented by the reference numerals are listed as follows:

[0047] 1-Industrial camera module, 2-Valve bag packaging machine, 3-Backlight module, 4-Telescopic cylinder. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] In the description of the present invention, it should be understood that the terms "right side", "upper left", "one side", "discharge port direction" and the like indicating directions or positional relationships are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the components or elements referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0050] A wrinkle detection method for a valve bag packaging machine of the present invention is implemented by a wrinkle detection device, which is installed on a valve bag packaging machine 2 and includes an industrial camera module 1, a backlight module 3 and a telescopic cylinder 4. The industrial camera module 1 is installed on the upper left of the valve bag packaging machine 2, the backlight module 3 is installed on the right side of the valve bag packaging machine 2, the backlight module 3 is vertically downward, and the telescopic cylinder 4 is installed on one side of the valve bag discharge port; Figure 3 As shown, for the wrinkle detection device, the specific real-time process is as follows: after the valve bag packaging machine 2 completes product packaging, the conveyor belt device moves toward the discharge port, and the industrial camera module 1 collects images and makes judgments. If a product with wrinkles in the packaging is detected, the telescopic cylinder 4 is started to remove the product;

[0051] The wrinkle detection method comprises the following steps:

[0052] Step 1: During the operation of the valve bag packaging machine 2, the backlight module 3 is adjusted so that the image texture of the valve bag product is clearly displayed on the display;

[0053] Step 2: Start the linear array industrial camera module 1 to collect the valve bag product image in real time and determine whether the image clarity meets the initial setting requirements;

[0054] Step 3: If the image clarity does not meet the initial setting requirements, the acquired image is enhanced and the enhanced image is output;

[0055] Step 4: If the image clarity meets the initial setting requirements, the collected image is subjected to YOLO forward inference to output the image wrinkle position information;

[0056] Step 5: Use the region segmentation and merging method to segment the image in the target frame, estimate the area based on the pixels, and quantitatively evaluate the wrinkle information;

[0057] Step 6: Establish the Kalman filter equation, track and detect products, and remove unqualified products on the conveyor line;

[0058] Step 7: Using the timestamp as the basic unit, record the work product data in real time and mark the work data where wrinkles appear on the seal;

[0059] Step 8: Input LSTM for network learning, and perform auxiliary adjustment on the production parameters of the packaging machine based on the output feedback information.

[0060] Among them, the image clarity specifically uses the average gradient as the image evaluation index. The average gradient can reflect the detail contrast and texture transformation in the image. The specific calculation is as follows:

[0061]

[0062] In the formula, ΔxF(i,j) and ΔyF(i,j) represent the first-order difference of pixel point (i,j) in the x or y direction respectively. By calculating a single product picture of a finished packaging bag, the clarity of the collected picture is evaluated to ensure that the texture features of the input image are clear.

[0063] Among them, the YOLO forward reasoning in step 4 includes the training of the YOLOX-S detection model for valve pocket wrinkles. By manually collecting some images and collecting an open source dataset, namely the valve pocket aluminum sheet surface defect detection dataset, 420 images are obtained. The wrinkle dataset is cyclically generated using style transfer to balance the number of positive and negative samples. The existing sample dataset is then expanded through geometric rotation, local splicing, and mirror flipping. The images are annotated using LabelImage software, and the annotated categories are divided into three categories: normal products, surface wrinkles, and sealing wrinkles. A wrinkle dataset is established and divided into a training set and a test set with a division ratio of 7:3.

[0064] Among them, the YOLO forward reasoning of the image in step 4 specifically adopts the YOLOX network structure of the improved YOLOX-S detection model, which mainly includes: removing the 13×13 feature map output from the three branches in the YOLO-Head part of the YOLOX-S detection model, which is responsible for the feature map of large targets, and introducing an attention mechanism module, and using a deep separable convolution operation to replace the ordinary convolution operation in the basic module.

[0065] Among them, the image is subjected to YOLO forward reasoning by inputting the data set into the YOLOX-S detection model in batches, inputting images with a width and height of 416×416×3, outputting feature layers of 26×26×8 and 52×52×8, and the length and width of the feature layer are 20, as well as the coordinate information (x, y, w, h) of the target boxes of three categories for prediction. Through the Reshape operation, the prediction box information of 3380×8 is obtained, and the losses are calculated respectively through back propagation, which mainly includes three parts: bounding box loss function, Iou perception classification loss and category loss. Since the wrinkles of the packaging are too small to detect the target in the collected image, the real box and the predicted box are often in an inclusive relationship. Therefore, the CIOU loss function is used as the bounding box loss function, and the CIOU loss function is used as the bounding box loss function; replace the original loss function, and the specific formula of the CIOU loss function is as follows:

[0066]

[0067]

[0068]

[0069] Among them, IoU is the intersection over union ratio of the predicted box and the real box, ρ represents the Euclidean distance between the two points, b and b gt They represent the center points of the predicted box and the real box respectively, c represents the diagonal distance between the predicted box and the minimum bounding rectangle of the real box, α is used as the coordination ratio parameter, γ is used to measure the length ratio consistency parameter of the box, and w gt , w and h gt , h represent the width and height of the real box and the predicted box respectively.

[0070] Among them, by setting the sparsity factor γ, the trained model is weighted and sorted, the maximum and minimum values ​​of the weights are removed, and the number of some convolution kernels in the middle convolution layer of the model is reduced through unstructured sparseness. The detection model is sparse, and the model is fine-tuned on the training data set to restore the accuracy of the model. Finally, the YOLOX-S detection model is deployed to the embedded terminal device.

[0071] Among them, in step five, the image is divided into a set of arbitrary non-overlapping initial areas, and these areas are split and merged according to the given uniformity detection criteria until the image is finally divided into the smallest number of uniform areas. After the image position of the product with wrinkles is determined, it is necessary to estimate the actual size of the wrinkles through pixel points. At the same time, by calculating the area of ​​pixel points, the degree of wrinkles can be roughly divided into three types: severe, medium, and slight.

[0072]

[0073] Among them, Sg is the pixel area of ​​the rectangular frame detected by YOLOX, and S is the pixel area after segmentation.

[0074] Among them, in step 6, the Kalman filter equation is established for target tracking with wrinkle problem to complete the update of target position. Since there is no additional control input during the observation process, uk is zero. Since the system has no additional control input, the state equation is:

[0075] x k =Ax k-1 +w k-1

[0076] x k =(x,y,v x ,v y ) T

[0077]

[0078] The coordinates of the target wrinkle are (x, y). Assume that the position of the detected object in the image changes in a uniform linear motion, and the change speeds are v and x , v y , A is the state transfer matrix, w k-1 is the process noise; the observation equation is:

[0079] z k =Hx k +v k

[0080]

[0081] z k is the observed value measured by YOLOX, the coordinates of the center point of the target in the image, H is the observation matrix, v k Represents the actual noise in the measurement process, and completes the update of the target position through prediction.

[0082] Among them, in step eight, the LSTM neural network is used to adjust the parameters of the equipment, mainly by inputting the working current of the packaging machine, the temperature at the sealing point, as well as the production rhythm of the product, the feeding rate, the actual quality of the packaged product and other actual control parameters, so as to assist in adjusting the various working parameters of the coating machine. Specific embodiment 1:

[0084] like Figure 1 As shown, a wrinkle detection method for a valve bag packaging machine includes the following steps:

[0085] Step 101, during the operation of the packaging machine, adjust the backlight source so that the image texture is clearly displayed on the display.

[0086] Step 102, start the line array camera to collect images in real time, and determine whether the image clarity meets the initial setting requirements;

[0087] Step 103, if the image clarity does not meet the initial setting requirements, the captured image is enhanced and the enhanced image is output;

[0088] Step 104: If the image clarity meets the initial setting requirements, the collected image is subjected to YOLO forward reasoning to output image wrinkle position information;

[0089] Step 105, using a region segmentation and merging method to segment the image in the target frame, estimate the area based on the pixels, and quantitatively evaluate the wrinkle information;

[0090] Step 106, establishing a Kalman filter equation, tracking and inspecting products, and removing unqualified products on the conveyor line;

[0091] Step 107, using the timestamp as a basic unit, recording the work product data in real time, and marking the work data where wrinkles appear on the seal;

[0092] Step 108, input LSTM to perform network learning, and perform auxiliary parameter adjustment on the packaging machine production parameters according to the output feedback information. Specific embodiment 2:

[0094] like Figure 2 As shown,

[0095] A wrinkle detection method for a valve bag packaging machine is disclosed, and the advantages of the present invention compared with the prior art are described by taking an experimental example.

[0096] (1) Start the packaging machine and set various production parameters.

[0097] (2) Ensure that the packaging machine is working by checking the material device.

[0098] (3) Adjust the backlight source and capture images using a linear array camera.

[0099] (4) Evaluate the image clarity. If the effect is poor, perform image enhancement before inputting the image.

[0100] (5) Input the image into the YOLOX target detection network, perform packaging bag wrinkle detection, and obtain the target detection result.

[0101] (6) Read the inspection category. If wrinkles appear on the surface and seal, track the wrinkle target and remove the product at the conveyor line.

[0102] (7) At the same time, the images of normal products, products with surface wrinkles, and products with sealing wrinkles are segmented, the wrinkle conditions are quantitatively evaluated, and then the product data is recorded.

[0103] (8) Integrate various production parameters and quantitative evaluation results, input them into the LSTM network model, establish a work prediction model, and assist in adjusting production parameters.

[0104] The present invention inputs the collected images into the YOLO-based deep learning detection model, and segments and identifies the identified foreign body images. The lightweight detection model effectively provides an online detection solution and quickly eliminates unqualified products. The technical means of segmenting the target image is adopted to realize the quantitative evaluation of the wrinkle information, thereby improving the sorting automation level of wrinkle detection, reducing the labor intensity of workers, and ensuring the accuracy of wrinkle detection.

[0105] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A wrinkle detection method for a valve bag packaging machine, implemented by a wrinkle detection device, the wrinkle detection device is installed on the valve bag packaging machine (2), comprising an industrial camera module (1), a backlight module (3) and a telescopic cylinder (4), the industrial camera module (1) is installed on the upper left of the valve bag packaging machine (2), the backlight module (3) is installed on the right side of the valve bag packaging machine (2), the backlight module (3) is vertically downward, and the telescopic cylinder (4) is installed on one side of the valve bag discharge port, characterized in that: The wrinkle detection method comprises the following steps: Step 1: During the operation of the valve bag packaging machine (2), the backlight module (3) is adjusted so that the image texture of the valve bag product is clearly displayed on the display; Step 2: Start the linear array industrial camera module (1) to collect the image of the valve bag product in real time, and determine whether the image clarity meets the initial setting requirements; Step 3: If the image clarity does not meet the initial setting requirements, the acquired image is enhanced and the enhanced image is output; Step 4: If the image clarity meets the initial setting requirements, the collected image is subjected to YOLO forward inference to output the image wrinkle position information; Step 5: Use the region segmentation and merging method to segment the image in the target frame, estimate the area based on the pixels, and quantitatively evaluate the wrinkle information; Step 6: Establish the Kalman filter equation, track and detect products, and remove unqualified products on the conveyor line; Step 7: Using the timestamp as the basic unit, record the work product data in real time and mark the work data where wrinkles appear on the seal; Step 8: Input LSTM for network learning, and assist in adjusting the production parameters of the packaging machine according to the output feedback information; The image clarity specifically uses the average gradient as an image evaluation index. The average gradient can reflect the detail contrast and texture transformation in the image. The specific calculation is as follows: In the formula, ΔxF(i,j) and ΔyF(i,j) represent the first-order difference of pixel point (i,j) in the x or y direction respectively. By calculating a single product picture of a finished packaging bag, the clarity of the collected picture is evaluated to ensure that the texture features of the input image are clear. The YOLO forward reasoning in the step 4 includes the training of the YOLOX-S detection model for valve pocket wrinkles. By manually collecting some images and collecting an open source data set, namely, a valve pocket aluminum sheet surface defect detection data set, N images are obtained. The wrinkle data set is cyclically generated by style transfer, and the number of positive and negative samples is balanced. Then, the existing sample data set is expanded by geometric rotation, local splicing, and mirror flipping. The images are annotated by LabelImage software, and the annotated categories are divided into three types of targets: normal products, surface wrinkles, and sealing wrinkles. A wrinkle data set is established, and the data set is divided into a training set and a test set with a division ratio of 7:

3. In the step 4, the image is subjected to YOLO forward reasoning by using a YOLOX network structure of an improved YOLOX-S detection model, which mainly includes: removing the 13×13 feature map outputted from the three branches in the YOLO-Head part of the YOLOX-S detection model, which is responsible for the feature map of large targets, and introducing an attention mechanism module, and replacing the ordinary convolution operation in the basic module with a deep separable convolution operation; The image is subjected to YOLO forward reasoning by inputting the data set into the YOLOX-S detection model in batches, inputting images with a width and height of 416×416×3, outputting feature layers of 26×26×8 and 52×52×8, and the length and width of the feature layer are 20, as well as the coordinate information (x, y, w, h) of the three categories of target boxes for prediction. Through the Reshape operation, the prediction box information of 3380×8 is obtained, and the losses are calculated respectively through back propagation, which mainly includes three parts: the bounding box loss function, the Iou perception classification loss and the category loss. The CIOU loss function is used as the bounding box loss function; the original loss function is replaced, and the specific formula of the CIOU loss function is as follows: Among them, IoU is the intersection over union ratio of the predicted box and the real box, ρ represents the Euclidean distance between the two points, b and b gt They represent the center points of the predicted box and the real box respectively, c represents the diagonal distance between the predicted box and the minimum bounding rectangle of the real box, α is used as the coordination ratio parameter, γ is used to measure the length ratio consistency parameter of the box, and w gt , w and h gt , h represent the width and height of the real box and the predicted box respectively; By setting the sparsity factor γ, the trained model is weighted and the maximum and minimum weights are removed. The number of some convolution kernels in the middle convolution layer of the model is reduced through unstructured sparsity, the detection model is sparse, and the model is fine-tuned on the training data set to restore the accuracy of the model. Finally, the YOLOX-S detection model is deployed to the embedded terminal device. In the step 5, the image is divided into a set of random non-intersecting initial regions, and these regions are split and merged according to a given uniformity detection criterion until the image is finally divided into the least number of uniform regions. After the image position of the product with wrinkles is determined, the actual size of the wrinkles needs to be estimated by pixel points. At the same time, the wrinkle degree is divided into three types: severe, medium, and slight by calculating the area of ​​the pixel points. In step six, the Kalman filter equation is established to track the target with wrinkle problem, and the target position is updated; In the step eight, the LSTM neural network is used to adjust the parameters of the equipment, mainly by inputting the working current of the packaging machine, the temperature at the sealing point, the production rhythm of the product, the feeding rate, and the actual control parameters of the packaged product, so as to assist in adjusting the various working parameters of the coating machine.

Images