Deep learning-based paper tube defect identification method and system

Abstract

The invention relates to the technical field of spandex paper tube quality detection, in particular to a paper tube defect identification method and system based on deep learning. The method comprises the following steps: acquiring a multi-azimuth image of a spandex paper tube, dividing a buffer area and a target area in the image, and respectively identifying defect conditions in the buffer area and the target area; integrating the buffer regions of the adjacent azimuth images, obtaining a fusion region, updating defect data in the fusion region, and outputting current spandex paper tube defect data; aiming at the defect data, identifying the defect grade of the current spandex paper tube, and outputting operation suggestion data; the system comprises a regional defect identification module, a regional defect fusion module and a defect grade identification module. By dividing the buffer area and the target area of the spandex paper tube image and integrating the buffer areas of the adjacent orientation images, the defect data of the spandex paper tube is obtained, and the defect level is identified, so that the accuracy and reliability of defect detection are improved.

Description

Technical Field

[0001] This invention relates to the field of spandex paper tube quality inspection technology, and in particular to a paper tube defect identification method and system based on deep learning. Background Technology

[0002] During the production of spandex paper tubes, various defects may occur due to factors such as raw material quality and production processes, including surface scratches, dents, cracks, and uneven coloring. These defects not only affect the appearance quality of the paper tubes but may also reduce their strength and performance, thus impacting the winding effect of the spandex filaments and overall product quality. Traditional methods for detecting paper tube defects mainly rely on manual visual inspection. This method suffers from drawbacks such as low efficiency, high labor intensity, and significant susceptibility to subjective factors, making it difficult to meet the demands of modern large-scale production.

[0003] With the development of computer vision and deep learning technologies, automatic defect detection methods based on image processing have gradually become a research hotspot. However, existing image processing-based detection methods, when processing images of spandex paper tubes, require the acquisition of image data from multiple directions because the paper tubes are cylindrical. This can lead to overlap of image data from adjacent directions, resulting in poor accuracy and reliability of defect detection. Summary of the Invention

[0004] The purpose of this invention is to provide a paper tube defect identification method and system based on deep learning, which aims to solve the technical problem in the prior art that when processing images of spandex paper tubes, since the paper tubes are cylindrical, it is necessary to collect image data from multiple directions of the spandex paper tubes. The image data from adjacent directions will overlap, resulting in poor accuracy and reliability of defect detection.

[0005] To achieve the above objectives, this invention employs a deep learning-based method for identifying paper tube defects, comprising the following steps: Acquire multi-angle images of spandex paper tubes, and divide the images into buffer areas and target areas, and identify the defects in the buffer areas and target areas respectively; Integrate the buffer areas of adjacent orientation images, obtain the fused region, update the defect data in the fused region, and output the current spandex paper tube defect data; Based on the defect data, identify the current defect level of the spandex paper tube and output operational suggestion data.

[0006] In the steps of acquiring multi-angle images of spandex paper tubes, dividing the images into buffer areas and target areas, and identifying defects in the buffer areas and target areas respectively: Establish detection points, trigger multi-angle camera shooting requests, acquire multi-angle images of spandex paper tubes, and preprocess the multi-angle images; Obtain the buffer region and target region in the image; Output buffer area and target area defect type and defect severity information.

[0007] In the steps of obtaining the buffer region and the target region in the image: Define the rules for dividing the buffer region and the target region, detect the edge contour of the spandex paper tube, divide the buffer region and the target region in the image, and output the result.

[0008] In the steps of outputting defect type and defect severity information in the output buffer area and target area: Defect features are extracted from the buffer region and the target region respectively, the dimensionality of the defect features is reduced, and the defect features are classified.

[0009] In the steps of integrating the buffer area of ​​adjacent orientation images, obtaining the fused area, updating the defect data in the fused area, and outputting the current spandex paper tube defect data: Based on feature points, the buffer regions of adjacent orientation images are stitched together, and geometric correction and brightness adjustment are performed on the images to obtain the fused region; The defects in the fused region are re-identified, and the defect identification results of the fused region are compared and fused with the defect identification results of the buffer region to update the defect data.

[0010] Before the steps of stitching together buffer regions of adjacent orientation images based on feature points, and performing geometric correction and brightness adjustment on the image to obtain the fused region: Based on the camera's placement and shooting angle, determine the correspondence between adjacent azimuth images and obtain the buffer area portion in the adjacent images.

[0011] The process includes re-identifying defects in the fused region, comparing and fusing the defect identification results of the fused region with those of the buffer region, and updating the defect data. The defect data of the target area and the updated merged area are integrated to form the current defect data of spandex paper tubes, and the defect data is output in a visual way.

[0012] Among the steps, in identifying the current defect level of the spandex paper tube based on the defect data and outputting operational suggestion data: Based on the defect data, the defect levels of the current spandex paper tubes are identified. Based on the defect level identification results, obtain operational suggestions.

[0013] After obtaining operational suggestions based on the defect level identification results: The operational suggestions will be output in both text and voice formats.

[0014] This invention also provides a paper tube defect identification system based on deep learning, including a region defect identification module, a region defect fusion module, and a defect level identification module; wherein: The regional defect identification module is used to acquire multi-angle images of spandex paper tubes, divide the images into buffer areas and target areas, and identify defects in the buffer areas and target areas respectively. The regional defect fusion module is used to integrate the buffer areas of adjacent orientation images, obtain the fused area, update the defect data in the fused area, and output the current spandex paper tube defect data. The defect level identification module is used to identify the current defect level of the spandex paper tube based on the defect data and output operation suggestion data.

[0015] This invention discloses a deep learning-based paper tube defect identification method and system, which employs a region defect identification module, a region defect fusion module, and a defect level identification module to perform the following steps: acquiring multi-directional images of spandex paper tubes and dividing the images into buffer regions and target regions, respectively identifying defects in the buffer regions and target regions; integrating buffer regions from adjacent directional images to obtain a fused region, updating defect data in the fused region, and outputting the current spandex paper tube defect data; identifying the current spandex paper tube defect level based on the defect data, and outputting operation suggestion data; by dividing the spandex paper tube images into buffer regions and target regions, and integrating buffer regions from adjacent directional images, acquiring spandex paper tube defect data, and identifying defect levels, the accuracy and reliability of defect detection are improved. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. 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.

[0017] Figure 1 This is a flowchart of the paper tube defect identification method based on deep learning according to the present invention.

[0018] Figure 2 This is a flowchart of steps S100 of the present invention.

[0019] Figure 3 This is a flowchart of steps S200 of the present invention.

[0020] Figure 4 This is a flowchart of steps S300 of the present invention.

[0021] Figure 5 This is a schematic diagram of the structural principle of the paper tube defect identification system based on deep learning of the present invention.

[0022] Figure 6 This is a schematic diagram of the electronic device of the present invention.

[0023] 401 - Regional Defect Identification Module, 402 - Regional Defect Fusion Module, 403 - Defect Level Identification Module. Detailed Implementation

[0024] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0025] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0026] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0027] Please see Figures 1-4 This invention provides a paper tube defect identification method based on deep learning, comprising the following steps: S100: Acquire multi-angle images of spandex paper tubes, divide the images into buffer areas and target areas, and identify defects in the buffer areas and target areas respectively.

[0028] In this embodiment, multi-angle images of the spandex paper tube are acquired, and the images are divided into buffer areas and target areas. Defects in the buffer areas and target areas are then identified respectively. The specific process is as follows: S101: Set up a detection point, trigger a multi-angle camera shooting request, acquire multi-angle images of spandex paper tubes, and preprocess the multi-angle images; S102: Define the rules for dividing the buffer area and the target area, detect the edge contour of the spandex paper tube, divide the buffer area and the target area in the image, and output the result; S103: Extract defect features from the buffer region and the target region respectively, reduce the dimensionality of the defect features, classify the defect features, and output the defect type and defect severity information of the buffer region and the target region.

[0029] In the above process, inspection points are set up at appropriate locations on the spandex paper tube production line. These inspection points should ensure that the appearance information of the paper tube can be clearly and comprehensively captured. Typically, this is chosen after the paper tube has completed its initial forming and its surface condition is relatively stable, such as after the paper tube has completed the winding and drying processes. Simultaneously, a stable lighting system is set up around the inspection point to ensure uniform illumination and avoid the impact of shadows and reflections on image quality.

[0030] Photoelectric sensors or other types of triggering devices are installed at the detection points. When the spandex paper tube arrives at the detection point, the triggering device detects its presence and sends a signal. This signal is simultaneously transmitted to a multi-directional camera array surrounding the paper tube, triggering all cameras to simultaneously capture images of the paper tube from different angles. The camera's shooting parameters, such as resolution, focal length, and exposure time, should be pre-adjusted and optimized according to the size of the paper tube and the detection requirements to ensure image clarity and integrity.

[0031] Grayscale conversion: Converting multi-view color images into grayscale images. Since color images contain information from three channels (red, green, and blue), the data volume is large, while grayscale images only have one channel, reducing the amount of data required for subsequent processing and improving processing speed. A weighted average method can be used for grayscale conversion.

[0032] Denoising: Median filtering is used to denoise the grayscale image. Median filtering is a non-linear filtering method that sorts the pixel values ​​in the neighborhood of each pixel in the image and takes the median as the new value of that pixel. It can effectively remove interference information such as salt-and-pepper noise in the image, while preserving the edge information of the image relatively well.

[0033] Contrast Enhancement: Histogram equalization is applied to the denoised image to enhance its contrast. Histogram equalization redistributes the gray values ​​of pixels in the image, making the gray-level distribution more uniform and thus making defect features more prominent. This can be implemented using relevant functions from image processing libraries such as OpenCV.

[0034] Based on the structural characteristics and testing requirements of spandex paper tubes, pre-define the rules for dividing the buffer area and the target area. For example, considering that the two ends of the paper tube may be affected differently during production, and that some areas at both ends may not directly participate in key functions in actual use, a certain length range (e.g., 5~10mm) at each end of the paper tube is divided into a buffer area, and the middle part is divided into a target area.

[0035] Edge contour detection is performed using the Canny edge detection algorithm to detect the edge contours of spandex paper tubes. First, the image is smoothed using Gaussian filtering to reduce noise. Then, the gradient magnitude and direction of the image are calculated to find edge points. Finally, non-maximum suppression and double threshold detection are used to determine the final edge contours.

[0036] Region segmentation and output: Based on detected edge contours and predefined segmentation rules, the buffer region and target region are accurately segmented in the image. Region segmentation can be implemented using image processing software or programming languages ​​(such as Python's OpenCV library). After segmentation, the results are output as an image, with the location and extent of the buffer and target regions marked on the image to facilitate subsequent defect identification and processing.

[0037] Defect feature extraction employs a deep learning-based approach to extract defect features from the buffer and target regions. A convolutional neural network (CNN) model is constructed, comprising an input layer, multiple convolutional layers, pooling layers, and fully connected layers. The pre-defined buffer and target region images are input into the CNN model. The convolutional layers perform convolution operations with the images using kernels to extract local image features. The pooling layers downsample the feature maps output by the convolutional layers, reducing the number of features and computational cost while enhancing feature robustness. Through alternating processing by multiple convolutional and pooling layers, a high-level feature representation of the image is obtained.

[0038] To reduce computational load and avoid overfitting, the extracted high-dimensional defect features are subjected to dimensionality reduction processing. Principal Component Analysis (PCA) can be used to project the original data into a new coordinate system through linear transformation, selecting the directions with the largest variance as the new feature dimensions, thereby reducing the feature dimensionality while retaining the main information of the data.

[0039] Defect feature classification uses Support Vector Machines (SVMs) or classification networks in deep learning (such as fully connected neural networks) to classify the dimensionality-reduced defect features. The dataset, labeled with defect types and severity, is divided into training and testing sets. The classification model is trained using the training set, adjusting its parameters to accurately distinguish between different types and degrees of defects. After training, the model is evaluated using the testing set to ensure high accuracy and generalization ability. Finally, the buffer region and target region images are input into the trained classification model, which outputs the defect type (e.g., scratches, dents, cracks, uneven coloring) and severity (minor, moderate, severe) for each region.

[0040] S200: Integrate the buffer areas of adjacent orientation images, obtain the fused area, update the defect data in the fused area, and output the current spandex paper tube defect data.

[0041] In this embodiment, buffer areas of adjacent orientation images are integrated to obtain a fused region, defect data in the fused region is updated, and the current spandex paper tube defect data is output. The specific process is as follows: S201: Based on the camera's position and shooting angle, determine the correspondence between adjacent orientation images and obtain the buffer area portion in the adjacent images; S202: Based on feature points, the buffer regions of adjacent orientation images are stitched together, and geometric correction and brightness adjustment are performed on the images to obtain the fused region; S203: Re-identify defects in the fused region, compare and merge the defect identification results of the fused region with the defect identification results of the buffer region, and update the defect data; S204: Integrate the defect data of the target area and the updated fusion area to form the current spandex paper tube defect data, and output the defect data in a visual manner.

[0042] In the above process, based on the camera's placement and shooting angle, the correspondence between adjacent azimuth images is determined, and the buffer area in adjacent images is obtained. Camera parameter recording involves accurately recording the placement and shooting angle information of each camera during installation. This information can be represented by parameters such as camera installation coordinates and rotation angles. For example, a three-dimensional coordinate system can be used to record the camera's position, and Euler angles can be used to record the camera's shooting direction.

[0043] The correspondence is determined by using geometric calculations based on the camera's position and shooting angle to establish relationships between adjacent images. For example, for images taken by two adjacent cameras, the correspondence can be determined by calculating the overlapping area between the two cameras. Feature point matching methods, such as the SIFT (Scale-Invariant Feature Transform) algorithm, can be used to extract feature points from the images and calculate the similarity between them to find corresponding feature points in adjacent images, thus establishing the correspondence between the images.

[0044] Buffer region extraction involves extracting the buffer region from adjacent images based on a defined correspondence. Image masking can be used to generate a mask image based on the location and extent of the buffer region. A bitwise AND operation is then performed between the mask image and the original image to obtain the buffer region image.

[0045] Feature point matching and stitching are performed using feature point matching algorithms such as SIFT or SURF (SpeededUpRobustFeatures) to extract feature points in the buffer regions of adjacent orientation images and calculate the matching relationships between feature points. Based on the matched feature point pairs, the Random Sample Consensus (RANSAC) algorithm is used to estimate the transformation matrix (such as an affine transformation matrix or a projective transformation matrix) between the images. Then, the buffer regions of adjacent images are stitched together according to the transformation matrix to obtain a preliminary stitched image.

[0046] Geometric correction addresses the issue that stitched images may exhibit geometric distortions, such as perspective distortion, due to the influence of camera shooting angle and position. Geometric correction algorithms are used to correct these distortions. Based on the image's transformation matrix, geometric correction uses a reverse mapping method to map pixels in the stitched image to their correct positions, resulting in a geometrically corrected image.

[0047] Brightness adjustment is crucial because images captured by different cameras may exhibit brightness differences, affecting the quality of the stitched image. Methods such as histogram matching and brightness equalization are used to adjust the brightness of the stitched image, ensuring uniform brightness. Histogram matching adjusts brightness by matching the histogram of one image to the histogram of another; brightness equalization applies linear or non-linear transformations to the pixel values ​​of the images to achieve a more even brightness distribution. After geometric correction and brightness adjustment, the fused region image is obtained.

[0048] For defect identification in the fused region, the same defect identification method as in step S103 is used to identify defects in the fused region image. The fused region image is input into a trained CNN model for feature extraction, and then a classification model is used to classify the extracted features to obtain information on the defect type and degree of defect in the fused region.

[0049] The results comparison and fusion process compares the defect identification results of the fused region with the results of the previous individual identification of the buffer region. The comparison includes aspects such as defect type and defect severity. If new defects are found in the fused region or the assessment of existing defects is more accurate, the defect data is updated. For example, if a crack defect was not found when the buffer region was identified individually, but is found in the fused region, the crack defect information is added to the defect data; if the severity of a scratch defect is assessed as severe in the fused region, while it was previously assessed as moderate, the scratch defect severity is updated to severe.

[0050] The defect data of the target area and the updated merged area are integrated to form the current defect data of spandex paper tubes, and the defect data is output in a visual way.

[0051] Defect data integration combines the defect data from the target area with the updated merged area to form complete defect data for the current spandex paper tube. Defect data can include information such as defect type, location, and severity. Data structures (such as dictionaries or lists) can be used to store and organize this defect data. The integrated defect data is output visually for easy viewing and processing by operators. Images of the spandex paper tube can be displayed on a screen, with the location and type of defects labeled on the image, along with detailed textual information such as defect severity. Reports can also be generated from the defect data and output as printed or electronic documents.

[0052] S300: Based on the defect data, identify the current defect level of the spandex paper tube and output operation suggestion data.

[0053] In this embodiment, based on the defect data, the current defect level of the spandex paper tube is identified, and operational suggestion data is output. The specific process is as follows: S301: Based on the defect data, identify the level of defects in the current spandex paper tube; S302: Obtain operational suggestions based on the defect level identification results; S303: Output operation suggestion data in text and voice formats.

[0054] In the above process, a defect classification standard is established. A standard for classifying the defects of spandex paper tubes is pre-established, classifying defects into different levels based on their type and severity, such as minor defects, general defects, and severe defects. For example, for scratches, scratches less than 2mm in length and shallow in depth can be classified as minor defects; scratches between 2 and 5mm in length or deeper in depth can be classified as general defects; and scratches greater than 5mm in length or that significantly affect the paper tube structure can be classified as severe defects.

[0055] Defect level identification involves classifying the defects in the current spandex paper tubes according to established defect level classification standards. Each defect in the data is traversed, and based on its type and severity, it is categorized into the corresponding defect level. For example, if a 3mm long scratch exists in the data, it is classified as a general defect according to the standards.

[0056] Operational recommendations are formulated based on different defect levels. For example, paper tubes with minor defects can be repaired and reused; those with general defects can be sold as substandard products; and those with severe defects should be scrapped. Furthermore, certain special types of defects, such as cracks that affect the strength of the paper tube, should be scrapped regardless of their severity.

[0057] Operational recommendations are obtained based on the defect level identification results and according to the corresponding operational recommendation rules. For example, if the current spandex paper tube is determined to have a general defect, the operational recommendation obtained according to the rules is to downgrade its sale. The operational recommendation data is displayed on the screen or output to the printing device in text form. The text content should be clear and unambiguous, such as "This spandex paper tube has a general defect, and downgrading its sale is recommended."

[0058] Voice output uses speech synthesis technology to convert operation suggestions into speech signals, which are then played back through a speaker. Voice output allows operators to easily obtain operation suggestions when they are busy or unable to view the screen. Existing speech synthesis libraries (such as Baidu Speech Synthesis and iFlytek Speech Synthesis) can be used to implement the voice output function.

[0059] In this invention, firstly, multi-angle images of spandex paper tubes are acquired, and buffer areas and target areas are divided in the images. Defect conditions in the buffer areas and target areas are identified respectively. Then, buffer areas of adjacent images are integrated to obtain a fused area. Defect data in the fused area is updated, and the current defect data of the spandex paper tube is output. Finally, based on the defect data, the current defect level of the spandex paper tube is identified, and operation suggestion data is output. By dividing the buffer areas and target areas of the spandex paper tube images and integrating the buffer areas of adjacent images, defect data of the spandex paper tube is acquired, and the defect level is identified, thereby improving the accuracy and reliability of defect detection.

[0060] Corresponding to the aforementioned embodiments of the deep learning-based paper tube defect identification method, this application also provides embodiments of a deep learning-based paper tube defect identification system.

[0061] Figure 5 This is a block diagram illustrating a deep learning-based paper tube defect identification system according to an exemplary embodiment. (Refer to...) Figure 5 The system may include: a regional defect identification module 401, a regional defect fusion module 402, and a defect level identification module 403; wherein: The regional defect identification module 401 is used to acquire multi-angle images of spandex paper tubes, divide the images into buffer areas and target areas, and identify the defect situation in the buffer areas and target areas respectively. The regional defect fusion module 402 is used to integrate the buffer areas of adjacent orientation images, obtain the fusion area, update the defect data in the fusion area, and output the current spandex paper tube defect data. The defect level identification module 403 is used to identify the current defect level of the spandex paper tube based on the defect data and output operation suggestion data.

[0062] In this embodiment, the regional defect identification module 401 acquires multi-directional images of the spandex paper tube and divides the images into buffer areas and target areas, respectively identifying defects in the buffer areas and target areas; the regional defect fusion module 402 integrates the buffer areas of adjacent directional images, acquires the fused area, updates the defect data in the fused area, and outputs the current spandex paper tube defect data; the defect level identification module 403 identifies the current defect level of the spandex paper tube based on the defect data and outputs operation suggestion data; by dividing the spandex paper tube image into buffer areas and target areas, and integrating the buffer areas of adjacent directional images, the defect data of the spandex paper tube is acquired, and the defect level is identified, thereby improving the accuracy and reliability of defect detection.

[0063] Regarding the system in the above embodiments, the specific manner in which each module performs its operations has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0064] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0065] Accordingly, this application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; and, when the one or more programs are executed by the one or more processors, causing the one or more processors to implement the deep learning-based paper tube defect identification method described above. Figure 6 The diagram shown is a hardware structure diagram of any device with data processing capabilities, which is an embodiment of the deep learning-based paper tube defect identification system provided by the present invention. (Except for...) Figure 6 In addition to the processor, memory, and network interface shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0066] Accordingly, this application also provides a computer-readable storage medium storing computer instructions thereon, which, when executed by a processor, implement the deep learning-based paper tube defect identification method described above. The computer-readable storage medium can be an internal storage unit of any data processing device as described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0067] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0068] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A method for identifying paper tube defects based on deep learning, characterized in that, Includes the following steps: Acquire multi-angle images of spandex paper tubes, and divide the images into buffer areas and target areas, and identify the defects in the buffer areas and target areas respectively; Integrate the buffer areas of adjacent orientation images, obtain the fused region, update the defect data in the fused region, and output the current spandex paper tube defect data; Based on the defect data, identify the current defect level of the spandex paper tube and output operational suggestion data.

2. The paper tube defect identification method based on deep learning as described in claim 1, characterized in that, In the steps of acquiring multi-angle images of spandex paper tubes, dividing the images into buffer areas and target areas, and identifying defects in the buffer areas and target areas respectively: Establish detection points, trigger multi-angle camera shooting requests, acquire multi-angle images of spandex paper tubes, and preprocess the multi-angle images; Obtain the buffer region and target region in the image; Output buffer area and target area defect type and defect severity information.

3. The paper tube defect identification method based on deep learning as described in claim 2, characterized in that, In the steps of acquiring the buffer region and the target region in the image: Define the rules for dividing the buffer region and the target region, detect the edge contour of the spandex paper tube, divide the buffer region and the target region in the image, and output the result.

4. The paper tube defect identification method based on deep learning as described in claim 2, characterized in that, In the steps of outputting defect type and defect severity information for the output buffer area and target area: Defect features are extracted from the buffer region and the target region respectively, the dimensionality of the defect features is reduced, and the defect features are classified.

5. The paper tube defect identification method based on deep learning as described in claim 2, characterized in that, In the steps of integrating the buffer area of ​​adjacent orientation images, obtaining the fused area, updating the defect data in the fused area, and outputting the current spandex paper tube defect data: Based on feature points, the buffer regions of adjacent orientation images are stitched together, and geometric correction and brightness adjustment are performed on the images to obtain the fused region; The defects in the fused region are re-identified, and the defect identification results of the fused region are compared and fused with the defect identification results of the buffer region to update the defect data.

6. The paper tube defect identification method based on deep learning as described in claim 5, characterized in that, Before the steps of stitching together buffer regions of adjacent orientation images based on feature points, and performing geometric correction and brightness adjustment on the image to obtain the fused region: Based on the camera's placement and shooting angle, determine the correspondence between adjacent azimuth images and obtain the buffer area portion in the adjacent images.

7. The paper tube defect identification method based on deep learning as described in claim 6, characterized in that, After re-identifying defects in the fused region, comparing and fusing the defect identification results of the fused region with those of the buffer region, and updating the defect data: The defect data of the target area and the updated merged area are integrated to form the current defect data of spandex paper tubes, and the defect data is output in a visual way.

8. The paper tube defect identification method based on deep learning as described in claim 1, characterized in that, In the steps of identifying the current defect level of spandex paper tubes based on defect data and outputting operational suggestion data: Based on the defect data, the defect levels of the current spandex paper tubes are identified. Based on the defect level identification results, obtain operational suggestions.

9. The paper tube defect identification method based on deep learning as described in claim 8, characterized in that, After obtaining operational suggestions based on the defect level identification results: The operational suggestions will be output in both text and voice formats.

10. A deep learning-based paper tube defect identification system, applied to the deep learning-based paper tube defect identification method as described in claim 1, characterized in that, It includes a regional defect identification module, a regional defect fusion module, and a defect level identification module; among which: The regional defect identification module is used to acquire multi-angle images of spandex paper tubes, divide the images into buffer areas and target areas, and identify defects in the buffer areas and target areas respectively. The regional defect fusion module is used to integrate the buffer areas of adjacent orientation images, obtain the fused area, update the defect data in the fused area, and output the current spandex paper tube defect data. The defect level identification module is used to identify the current defect level of the spandex paper tube based on the defect data and output operation suggestion data.

Images