Disposable absorbent article defect detection system and method based on AI vision

Abstract

The invention discloses a disposable absorbent article defect detection system and method based on AI vision, the system is arranged corresponding to a production line, an image acquisition module in a multi-modal imaging unit is arranged corresponding to a station on the production line so as to acquire an image and preprocess the image, the image acquisition module is arranged corresponding to an encoder, and the encoder is arranged corresponding to the multi-modal imaging unit. The encoder sends out a pulse signal to trigger the image acquisition modules, and all images corresponding to the same product acquired by all the image acquisition modules are associated based on the encoder; the AI processing unit is used for carrying out defect detection analysis and the like on the image; and the control unit is used for controlling the working state of the multi-mode imaging unit and forming a control instruction according to a detection and analysis result. According to the system, the purpose of seamless integration of defect information of multiple procedures across multiple paths and branches is achieved, so that the defects of the absorptive article can be detected more accurately and comprehensively, and products with the defects can be better removed in the follow-up process.

Description

Technical Field

[0001] This invention relates to the field of disposable hygiene product manufacturing technology, specifically to a defect detection system and method for disposable absorbent items based on AI vision. Background Technology

[0002] In the production process of disposable absorbent items such as disposable diapers or sanitary napkins, the production line typically includes steps such as laminating various layers of materials, folding continuous sheets, and cutting continuous sheets. To ensure product quality, existing technologies usually include rejection mechanisms that use detection systems to identify defective absorbent items and remove them from the production line. However, existing detection systems mainly rely on sensors to detect obvious defects (such as large stains and joints), but they cannot effectively identify some difficult-to-detect defects (such as hair, glue droplets, small stains, mixtures of glue and fibers, and hard foreign objects inside the product). Furthermore, the detection devices in existing systems are often located at the end of the production line, making it difficult to identify defects at different levels of materials during production, especially subtle defects or those with little color difference from the material itself, after being stacked with other layers. This can lead to defective products containing these defects entering the market, affecting product quality. In addition, the variety of defects makes them difficult to statistically analyze, posing challenges to quality control. Summary of the Invention

[0003] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0004] The present invention provides an AI vision-based defect detection system and method for disposable absorbent items to solve one or more of the above-mentioned technical problems.

[0005] An AI-based vision-based defect detection system for disposable absorbent items is installed correspondingly on the production line. The system includes a multimodal imaging unit, an AI processing unit, and a control unit. Both the multimodal imaging unit and the AI ​​processing unit are connected to the control unit, and the multimodal imaging unit is also connected to the AI ​​processing unit. The multimodal imaging unit is used to acquire images and preprocess them; The AI ​​processing unit is used to detect, classify, statistically analyze, and generate detection results from the preprocessed image, which include the type, location, and number of defects. The control unit is used to control the working state of the multimodal imaging unit and generate control commands based on the detection results.

[0006] In some embodiments, the multimodal imaging unit includes The image acquisition module group contains one or more image acquisition modules. The image acquisition modules are set up to correspond to the workstations on the production line that perform different technological operations. The image acquisition modules acquire images of the products at the corresponding workstations, and these images are the original images. The preprocessing module is used to preprocess the original image, transforming it into an image suitable for subsequent tasks.

[0007] In some implementations... The image acquisition module includes a lens, a camera, a light source, and an image acquisition card. The camera is a line scan camera, and the light source is a linear ELD light source that flashes synchronously with the line scan camera. The preprocessing process includes noise removal, contrast enhancement, correction of uneven lighting, sharpening, size normalization, grayscale / color normalization, geometric correction, pixel normalization, downsampling, and ROI extraction.

[0008] In some implementations, the image acquisition module uses polarizing filters or multi-angle light sources to eliminate reflections, and multiple sets of cameras are arranged at different workstations on the production line to ensure that no product defects are missed.

[0009] In some implementations... The production line is equipped with an encoder, which is connected to an image acquisition module. The encoder emits pulse signals, and the image acquisition module scans and takes pictures based on the pulse signals. The encoder associates images of the same product acquired by all image acquisition modules. If a defect is found in the detection results of any one or more of the associated images, the control unit generates a corresponding rejection command.

[0010] In some implementations... Whenever the camera is triggered to capture a line of images, the current encoder count value is latched and used as a timestamp or location stamp to bind the line of image data, thereby achieving the purpose of associating images of the same product captured by all image acquisition modules; The production line is equipped with a PLC industrial control computer, which is connected to the control unit. The control unit forwards the control instructions generated based on the detection results to the PLC industrial control computer, which then controls the corresponding products on the production line to be removed based on the rejection instructions.

[0011] In some implementations, the AI ​​processing unit includes a pre-trained deep learning model and a data preprocessing module, wherein The deep learning model is used to analyze the line flow of images acquired by each camera in real time after preprocessing. The model also identifies and outputs defect information, including the type of defect, the location of the defect in the image (pixel coordinates x along the scanning direction, start line number Y_start and end line number Y_end along the motion direction), defect size (size / contour), confidence level, etc. The data preprocessing module includes statistical analysis software, which classifies and statistically analyzes the received defect reports, counts the number of defects corresponding to the associated images, and forms the detection results.

[0012] In some implementations, the deep learning model also has a coordinate transformation function, which transforms the position of the defect in the image to a global absolute position space. The specific transformation process is as follows: When a defect is detected in an image captured by a camera, the location of the defect in the image is reported (Y_start_local, Y_end_local). Read the latch encoder values ​​(Enc_start, Enc_end) associated with the start and end lines corresponding to the defect. Using the camera's calibration parameters (Offset, Scale_Factor), calculate the defect's starting point Global_Start and ending point Global_End in the global absolute position space: Global_Start=Enc_start+Offset*Scale_Factor, Global_End=Enc_end+Offset*Scale_Factor.

[0013] The AI-based vision-based defect detection method for disposable absorbent items applies the aforementioned AI-based vision-based defect detection system for disposable absorbent items. The specific detection method includes: Image acquisition; Image preprocessing; Defect detection involves performing defect detection on preprocessed images to identify the type, location, and number of defects. Image classification and defect statistics classify images corresponding to the same product into one category and count the number of defects in images of the same category. If the result is 0, the product is determined to be defect-free; otherwise, the product is determined to be defective.

[0014] Compared with existing technologies, the beneficial effects of this invention are: the introduction of visual AI detection technology to detect various defects, including contaminants and joints of different shapes, sizes and colors, thereby improving the detection accuracy and efficiency of the production line, while providing more detailed data support for quality control. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and detailed embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a structural block diagram of a single-use absorbent item defect detection system based on AI vision in some embodiments of the present invention; Figure 2 This is a flowchart of a method for detecting defects in disposable absorbent items based on AI vision in some embodiments of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0017] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0018] Secondly, the present invention will be described in detail with reference to the schematic diagrams. When describing the embodiments of the present invention, for ease of explanation, the cross-sectional views illustrating the device structure will be partially enlarged, not according to the usual scale. Furthermore, the schematic diagrams are merely examples and should not limit the scope of protection of the present invention. In addition, actual fabrication should include the three-dimensional spatial dimensions of length, width, and depth.

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0020] Combination Figure 1 As shown in the embodiment, this invention proposes a defect detection system for disposable absorbent items based on AI vision, including a multimodal imaging unit, an AI processing unit, and a control unit. Both the multimodal imaging unit and the AI ​​processing unit are connected to the control unit, and the multimodal imaging unit is also connected to the AI ​​processing unit. The multimodal imaging unit is used to acquire images and preprocess them; The AI ​​processing unit is used to detect, classify, statistically analyze, and generate detection results from the pre-processed images, including the type, location, and number of defects. The control unit is used to control the working state of the multimodal imaging unit and generate control commands based on the detection results.

[0021] The aforementioned multimodal imaging unit may include The image acquisition module group contains one or more image acquisition modules. The image acquisition modules are set up to correspond to the workstations on the production line that perform different process operations. The image acquisition modules acquire images of the products (such as disposable absorbent items) at the corresponding workstations, and these images are the original images. The preprocessing module is used to preprocess the original image, transforming it into an image suitable for subsequent tasks (including detection, classification, statistics, etc.).

[0022] In some specific implementations, the multimodal imaging unit may also include a storage module for storing the original image and / or the preprocessed image.

[0023] An encoder is installed on the production line. This encoder connects to an image acquisition module via a control unit. The encoder emits pulse signals, and the corresponding image acquisition module scans and takes pictures based on these pulse signals. The encoder can be a high-precision incremental rotary encoder installed on the main drive roller of the production line (which is typically the roller that controls the rotation of the forming blades in the forming equipment). This encoder outputs one or more pulse signals with each rotation of the roller. Each pulse signal corresponds to a fixed physical distance (e.g., 0.1 mm, 0.05 mm, etc.) moved by the roller (i.e., the product on the production line). This correspondence determines the minimum positional accuracy of this AI vision-based disposable absorbent item defect detection system.

[0024] The production line is also equipped with a PLC industrial control computer, which is connected to the control unit. The control unit generates control commands based on the detection results and forwards them to the PLC industrial control computer. The PLC industrial control computer then controls the corresponding mechanisms on the production line (such as existing rejection mechanisms) to remove defective products from the production line. This process can be directly implemented using existing technology, so it will not be described in detail here.

[0025] The aforementioned image acquisition module may include a lens, a camera, a light source, and an image acquisition card. The camera may be a line scan camera, specifically a high-resolution industrial line scan camera (e.g., 5 megapixels or higher). Each line scan camera is equipped with an image acquisition card, which can receive pulse signals from the encoder.

[0026] Normally, whenever the encoder generates a certain number of pulses (e.g., N pulses), the image acquisition card triggers the camera to capture one line of pixels. This value N can be calculated and set based on the desired image resolution (along the motion direction) and the encoder resolution. Specifically, the triggering mechanism in this embodiment can be set as follows: The encoder's Zero Pulse / Index Pulse (issued once per revolution) is used as the line trigger reference, combined with pulse technology for precise inter-line positioning.

[0027] Whenever the camera is triggered to acquire a line of images, the current encoder count value is latched and used as a timestamp or location stamp, strictly bound to the image data of that line. This achieves the purpose of associating images of the same product acquired by all image acquisition modules. If a defect is found in the detection results of any one or more of the associated images, the control unit generates a corresponding rejection command.

[0028] The aforementioned light source can be a linear ELD light source, which flashes synchronously with the line scan camera to ensure that each line of the image is clear, without blurring or shadow interference. In addition, the image acquisition module can also use polarizing filters or multi-angle light sources to eliminate reflections, and multiple sets of cameras can be arranged at different workstations on the production line to ensure that no product defects are missed.

[0029] The above preprocessing process may include: Noise Removal: Eliminates random pixel interference introduced by sensors, transmission processes, or lighting conditions, resulting in clearer images; Enhance contrast: Stretch or adjust the grayscale / color range of the image to make the difference between target features and the background more obvious and the details more prominent; Correcting uneven lighting: Compensating for areas in the image that are too bright or too dark due to the position or occlusion of the light source; Sharpening: Enhances edges and details, making the image look sharper; Size normalization: Adjusting all images to a uniform size is crucial for subsequent algorithms (especially deep learning models), as they typically require fixed-size inputs; Grayscale / color normalization: Converting an image to a specific color space (e.g., grayscale, HSV, LAB) to simplify analysis or meet algorithm requirements; Geometric correction: Corrects geometric distortions caused by lens distortion (barrel and pincushion distortion) or shooting angle; Pixel normalization: Scales pixel intensity values ​​to a specific range (e.g., [0, 1] or [-1, 1]) to accelerate model convergence and improve numerical stability; Downsampling: Reduces image resolution (size), reduces the amount of data that needs to be preprocessed, and improves preprocessing speed; ROI extraction: preprocessing only the regions of interest in the image, either cropped or retained.

[0030] The aforementioned preprocessing module can transmit the preprocessed image to the AI ​​processing unit via high-speed Ethernet.

[0031] The aforementioned AI processing unit includes a pre-trained deep learning model (e.g., a deep learning-based image segmentation or object detection model) and a data preprocessing module, wherein... The deep learning model analyzes the pre-processed image line streams acquired by each camera in real time (which are then stitched together into a continuous image band based on the aforementioned relationships). The model can also identify defect information, generate a defect report, and output the defect information, including the defect type, the defect's location in the image (pixel coordinates x along the scanning direction, the start line number Y_start and the end line number Y_end along the motion direction), the defect size (size / contour), and confidence level. The data preprocessing module includes statistical analysis software, which classifies and statistically analyzes the received defect reports, counts the number of defects corresponding to the associated images, and forms the detection results.

[0032] The aforementioned control unit can be connected to the AI ​​processing unit via gigabit broadband or similar means.

[0033] The aforementioned control unit can monitor the images received by the AI ​​processing unit based on the encoder's synchronization signal, monitor frame loss and frame dropping, and issue an alarm command when frame loss or frame dropping is detected, so as to control the alarm-issuing device connected to the control unit to issue an alarm.

[0034] In some specific implementations, the encoder on the main drive roller can provide a unified position pulse, triggering image acquisition by all image acquisition modules. The position information corresponding to the position pulse is bound to each row of images for spatial calibration. The AI ​​vision-based disposable absorbent item defect detection system obtains the absolute physical position of each camera's field of view on the final product. When the AI ​​processing unit detects a defect in the images acquired by each camera, it uses calibration parameters to transform the position of the defect from the camera's local coordinate system to a global absolute position coordinate system based on the encoder. This ensures that defects from all sources and of all types can be mapped to the same axis representing the product's length, achieving seamless integration of defect information across multiple branches and processes.

[0035] In addition, the aforementioned system may also include a dynamic learning component, in which new defect samples are collected during system operation, and the deep learning model is updated periodically or in real time (through online learning or incremental learning). This process can be directly implemented using existing technologies, and therefore will not be elaborated upon here.

[0036] Physical calibration objects, such as calibration plates or actual products with known physical markings (e.g., crosshairs, specific images, such as pentagrams, etc.), can be added to the aforementioned production line, running throughout the entire line. The AI ​​vision-based defect detection system for disposable absorbent items initially performs spatial calibration, using physical calibration objects throughout the production line. These objects pass through the line, with all cameras simultaneously capturing images. The system analyzes the position of the same physical calibration object in the images from each camera. Since the position of each camera relative to the final product is fixed, but the material may have passed through multiple rollers (stretching, relaxing) or branch paths by the time it reaches that camera, a mapping relationship from local position to absolute position can be established. The specific calculation process is as follows: The offset of any point within the field of view of each camera between the camera's own "Camera LocalEncoder Space" and the "Global Encoder Space" defined by the encoder, and possible scaling factors (if the material is significantly stretched between cameras), are considered. Therefore, the formula for calculating the mapping relationship from local to absolute position can be: Global_Position=Camera_Local_Position+Camera_Offset*Scale_Factor, In the formula, Global_Position is the absolute position based on the encoder, Camera_Local_Position is the position calculated based on the encoder latched when the current camera is triggered, Camera_Offset is the displacement offset between the current camera and the encoder (unit: encoder count), and Scale_Factor is the potential scaling factor of the material on this path (usually close to 1, negligible in inelastic materials; in elastic materials or areas with large tension variations, it can be directly measured using existing counts). Specific Offset and Scale_Factor are stored for each camera (especially material cameras on branches of the production line).

[0037] Combination Figure 2 The content shown presents a detection method that can be applied to the aforementioned AI vision-based defect detection system for disposable absorbent items. The method specifically includes the following steps: S1. Image acquisition: The image acquisition module is triggered by the encoder pulse signal to acquire images and bind the images to the corresponding encoder count values. S2. Image preprocessing: Preprocess the image. S3. Defect Detection: The AI ​​processing unit performs defect detection and recognition operations on the preprocessed image. When a defect is identified, its position in the image is converted to the "global absolute position space." The specific conversion process is as follows: When a defect is detected in an image captured by a camera, the location of the defect in the image (Y_start_local, Y_end_local) is reported (where Y is the local location calculated based on the camera's latch encoder value; specifically, each time the camera is triggered to capture a line of images, the current encoder count value is latched and used as a timestamp or location stamp, and the start position of the defect can be recorded through the location stamp). The latch encoder values ​​(Enc_start, Enc_end) associated with the start and end lines corresponding to the defect are read. Using the camera's calibration parameters (Offset, Scale_Factor), calculate the defect's starting point Global_Start and ending point Global_End in the "global absolute position space": Global_Start=Enc_start+Offset*Scale_Factor, Global_End=Enc_end+Offset*Scale_Factor, For the camera at the last station on the production line (e.g., when the product is a disposable absorbent item, the product needs to be composited with multiple different layers of materials, and the last station is the position after all the materials corresponding to the product have been composited in sequence), its offset is usually very small (relative to the encoder position), and the Scale_Factor is usually 1 (or already taken into account the encoder measurement).

[0038] S4. Defect Classification and Data Statistics: The system sorts all defect events according to Global_Start (or Global_End) on the absolute position coordinate axis of the encoder, and then performs integrated views and / or cross-view correlation. Integrated view: The system can construct a "time axis" or "location axis" that represents the entire length of the product and accurately mark all detected defects on this axis at their corresponding physical locations; Cross-view correlation: If a defect (e.g., a material hole) is detected from different angles by multiple cameras (e.g., when the product is a disposable absorbent item, it needs to import different materials from different production line branches for lamination; in this case, multiple cameras include cameras set on the production line branches and the cameras corresponding to the materials after lamination) even though they report the same physical defect, because the positions are all converted to the "global absolute position space," the system can automatically identify that this is the same defect event. These reports can be merged or correlated to provide more comprehensive information about each level of the product (e.g., whether the defect is on the surface, absorbent layer, or bottom layer). If the number of defects on the same absolute position coordinate axis is 0, the product is determined to be defect-free; otherwise, the product is determined to be defective.

[0039] In some specific implementations, after the above detection methods are completed, defect handling is carried out. The specific defect handling process can be as follows: the AI ​​vision-based disposable absorbent item defect detection system can generate control instructions based on statistical results according to the control unit, and send the control instructions to the PLC industrial control computer corresponding to the production line. The PLC industrial control computer controls the corresponding mechanism on the production line according to the control instructions to remove the defective products detected from the production line.

[0040] In some specific implementations, when a defective product arrives at the downstream rejection mechanism (e.g., an air valve), the system checks if there are any defects within the product's global location range (the range from when the camera takes the picture to when the rejection mechanism is activated, which can be tracked in real time by an encoder). If a defect is found, a precise rejection signal is issued. Defect maps or reports are generated, showing the distribution of product defects by location and type, and providing quality reports for process monitoring, quality traceability, and equipment maintenance decisions. The resulting defect maps or reports help operators analyze the locations and cameras where defects are concentrated, and facilitate the identification of problematic processes (e.g., material supply issues in a specific branch, or an abnormality in a composite roller).

[0041] The above deep learning model training process may include the following steps: Data acquisition: Images of products from different processing steps on several production lines are acquired through the image acquisition unit, including images of normal products and images of various defective products; Data annotation: The collected images are annotated to mark the location and type of defects, and the data is compiled into a dataset. The dataset is then divided into three parts according to a pre-set ratio or randomly, which serve as the training set, test set, and validation set, respectively. Model training: Train the deep learning model using the training set and optimize the model's defect detection capabilities using the test set; Model validation: The model is validated using a validation set, and the model parameters of the deep learning model are adjusted to improve detection accuracy.

[0042] The types of defects mentioned above may include: structural defects (material holes, misalignment of the flow guide layer, displacement of the absorbent core, folds, skewing, and presence of fiber clumps), appearance defects (stains, indentations, damage, and rough edges), functional defects (missing adhesive strips, incorrect placement of Velcro, and missing items), and foreign object defects (adhesive droplets, adhesive lumps, hair, metal objects, and waste material from cutting).

[0043] The aforementioned undisclosed matters can be directly implemented using existing technologies, so they will not be elaborated upon here.

[0044] Although the present invention has been described above with reference to embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of the invention. In particular, as long as there is no structural conflict, the features in the disclosed embodiments can be combined with each other in any manner. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A defect detection system for disposable absorbent items based on AI vision, characterized in that, An AI-based vision-based defect detection system for disposable absorbent items is installed correspondingly on the production line. The system includes a multimodal imaging unit, an AI processing unit, and a control unit. Both the multimodal imaging unit and the AI ​​processing unit are connected to the control unit, and the multimodal imaging unit is also connected to the AI ​​processing unit. The multimodal imaging unit is used to acquire images and preprocess them; The AI ​​processing unit is used to detect, classify, statistically analyze, and generate detection results from the preprocessed image, which include the type, location, and number of defects. The control unit is used to control the working state of the multimodal imaging unit and generate control commands based on the detection results.

2. The AI ​​vision-based defect detection system for disposable absorbent items according to claim 1, characterized in that, The multimodal imaging unit includes The image acquisition module group contains one or more image acquisition modules. The image acquisition modules are set up to correspond to the workstations on the production line that perform different technological operations. The image acquisition modules acquire images of the products at the corresponding workstations, and these images are the original images. The preprocessing module is used to preprocess the original image, transforming it into an image suitable for subsequent tasks.

3. The AI ​​vision-based defect detection system for disposable absorbent items according to claim 2, characterized in that, The image acquisition module includes a lens, a camera, a light source, and an image acquisition card. The camera is a line scan camera, and the light source is a linear ELD light source that flashes synchronously with the line scan camera. The preprocessing process includes noise removal, contrast enhancement, correction of uneven lighting, sharpening, size normalization, grayscale / color normalization, geometric correction, pixel normalization, downsampling, and ROI extraction.

4. The AI ​​vision-based defect detection system for disposable absorbent items according to claim 3, characterized in that, The image acquisition module uses polarizing filters or multi-angle light sources to eliminate reflections, and multiple sets of cameras are arranged at different workstations on the production line to ensure that no product defects are missed.

5. The AI ​​vision-based defect detection system for disposable absorbent items according to claim 2, characterized in that, The production line is equipped with an encoder, which is connected to an image acquisition module. The encoder emits pulse signals, and the image acquisition module scans and takes pictures based on the pulse signals. The encoder associates images of the same product acquired by all image acquisition modules. If a defect is found in the detection results of any one or more of the associated images, the control unit generates a corresponding rejection command.

6. The AI ​​vision-based defect detection system for disposable absorbent items according to claim 5, characterized in that, Whenever the camera is triggered to capture a line of images, the current encoder count value is latched and used as a timestamp or location stamp to bind the line of image data, thereby achieving the purpose of associating images of the same product captured by all image acquisition modules; The production line is equipped with a PLC industrial control computer, which is connected to the control unit. The control unit forwards the control instructions generated based on the detection results to the PLC industrial control computer, which then controls the corresponding products on the production line to be removed based on the rejection instructions.

7. The AI ​​vision-based defect detection system for disposable absorbent items according to claim 6, characterized in that, The AI ​​processing unit includes a pre-trained deep learning model and a data preprocessing module, in which... A deep learning model is used to analyze the line flow of images acquired by each camera in real time after preprocessing. The model also identifies and outputs defect information, including the type of defect, the location of the defect in the image, the size of the defect, and the confidence level. The data preprocessing module includes statistical analysis software, which classifies and statistically analyzes the received defect reports, counts the number of defects corresponding to the associated images, and forms the detection results.

8. The AI ​​vision-based defect detection system for disposable absorbent items according to claim 7, characterized in that, Deep learning models also have coordinate transformation capabilities, which transform the location of defects in an image to a global absolute location space. The specific transformation process is as follows: When a defect is detected in an image captured by a camera, the location of the defect in the image is reported (Y_start_local, Y_end_local). Read the latch encoder values ​​(Enc_start, Enc_end) associated with the start and end lines corresponding to the defect. Using the camera's calibration parameters (Offset, Scale_Factor), calculate the defect's starting point Global_Start and ending point Global_End in the global absolute position space: Global_Start=Enc_start+Offset*Scale_Factor, Global_End=Enc_end+Offset*Scale_Factor.

9. A defect detection method for disposable absorbent items based on AI vision, characterized in that, The detection method using the AI ​​vision-based defect detection system for disposable absorbent items as described in claim 8 includes: Image acquisition; Image preprocessing; Defect detection involves performing defect detection on preprocessed images to identify the type, location, and number of defects. Image classification and defect statistics classify images corresponding to the same product into one category and count the number of defects in images of the same category. If the result is 0, the product is determined to be defect-free; otherwise, the product is determined to be defective.

Images