A lace cloth defect detection system based on a cloud platform and an edge gateway

Abstract

This invention provides a lace fabric defect detection system based on a cloud platform and edge gateway in the field of industrial internet technology. The system includes an edge gateway, a fabric production line, cameras, a cloud platform, and a host computer. The edge gateway connects to the fabric production line, cameras, cloud platform, and host computer. The edge gateway is used to create and train a defect detection model, which is then used to detect defects in the lace fabric. The detection results are sent to the host computer and cloud platform, and the fabric production line is controlled based on control commands from the cloud platform or host computer. The fabric production line is used to produce, transport, and display the lace fabric. The cameras are used to capture video of the fabric production line, providing video of the lace fabric. The cloud platform and host computer are used to store and display the defect detection results and remotely control the fabric production line. The advantages of this invention are: significantly improved accuracy, generalization ability, and timeliness of lace fabric defect detection.

Description

Technical Field

[0001] This invention relates to the field of industrial internet technology, and in particular to a lace fabric defect detection system based on a cloud platform and edge gateway. Background Technology

[0002] Surface defects are unavoidable during the production of lace fabric. Common surface defects include missed stitches, broken warp threads, broken weft threads, loose threads, and excess yarn. To ensure product quality, defect detection is necessary. Traditionally, this is done manually, with production workers inspecting the lace fabric on an assembly line. They manually adjust the production line speed, visually inspecting the target area. If a defect is found, the line is immediately stopped, and the type of defect is judged and marked based on experience. Detailed information about important defects, such as shape and texture, is recorded, or the lace fabric is repaired before the line continues inspection. Because human perception of color and shape is highly dependent on personal experience, manual inspection is difficult to be objective and consistent. Furthermore, manual inspection is labor-intensive, demanding, and has a high rate of false positives and false negatives. To overcome these problems, current technology typically uses machine vision to assist in the defect detection of lace fabric.

[0003] Traditional image processing-based visual inspection methods suffer from low recognition rates for defects such as missing needles and tears, as well as high false negative and false positive rates. Supervised machine learning visual inspection techniques face several challenges in factory applications: First, models rely on human experience and require well-labeled defect pattern data. The process of collecting and labeling samples is lengthy, and models often adapt to specific datasets. Training relies on manual labels, leading to unsatisfactory detection accuracy when the sample size is small, the distribution is imbalanced, or the detection target is small. Second, existing models have poor adaptability to new lace fabric patterns. Due to the wide variety of lace fabric patterns—some large factories have thousands—and more than ten types of defects, models have low detection rates for new patterns, and development cycles are long. Third, the system's real-time performance is weak, making it difficult to respond to the high demands of production line operations. Furthermore, it lacks collaborative feedback adjustment capabilities, resulting in limited value from defect data.

[0004] A search revealed that Chinese invention patent application number CN202210156089.5, filed on February 21, 2022, discloses a method and device for detecting surface defects in lace fabrics based on image simulation enhancement. This patent only uses artificially created simulated defect videos for training, and the control of defect detection quality depends on the quality of artificially simulated defects and image enhancement. It is difficult to guarantee the detection accuracy of multiple defects in different patterns. Furthermore, this method does not involve how to process defects after detection, or the control of fabric quality inspection devices.

[0005] Chinese invention patent application number CN201710735292.7, filed on August 24, 2017, discloses a method and system for online detection of fabric defects based on machine vision. This patent can identify and locate defects through image processing and machine learning algorithms, and is equipped with automatic marking equipment and a host computer. It can mimic the human eye inspection work and manual marking operation on the production line, minimizing the amount of manual work. However, this method is only applicable to more common fabric inspection scenarios and does not address the problems of lace fabrics, which have diverse patterns, are difficult to inspect, and have a high rate of missed and false detections by manual inspection and traditional vision methods.

[0006] Chinese invention patent application number CN202011508963.4, filed on December 18, 2020, discloses a production line error correction auxiliary system and method based on an edge gateway. This patent can capture video data of production line personnel's production operations and the time it takes for products to flow through the production line in real time, and determine whether the operations of production line personnel meet the requirements, thereby reducing the waste of working hours and factory resources caused by unnecessary actions. However, it does not involve the function of detecting defects in lace fabrics.

[0007] Therefore, how to provide a lace fabric defect detection system based on cloud platform and edge gateway to improve the accuracy, generalization ability and timeliness of lace fabric defect detection has become an urgent technical problem to be solved. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a lace fabric defect detection system based on a cloud platform and edge gateway, so as to improve the accuracy, generalization ability and timeliness of lace fabric defect detection.

[0009] The present invention is implemented as follows: a lace fabric defect detection system based on a cloud platform and an edge gateway, comprising at least one edge gateway, several fabric production lines, several cameras, at least one cloud platform, and at least one host computer; the edge gateway is connected to the fabric production line, cameras, cloud platform, and host computer respectively.

[0010] The edge gateway is used to create and train a defect detection model, and to detect defects in lace fabric through the defect detection model to obtain defect detection results. The defect detection results are then sent to the host computer and cloud platform, and the operation of the fabric production line is controlled based on the control instructions of the cloud platform or the host computer.

[0011] The fabric production line is used to produce, convey, and display lace fabric;

[0012] The camera is used to film the fabric production line to capture video of lace fabric.

[0013] The cloud platform is used to store and display defect detection results and to remotely control the operation of the fabric production line;

[0014] The host computer is used to store and display defect detection results and remotely control the operation of the fabric production line.

[0015] Furthermore, the creation and training of the defect detection model specifically involves:

[0016] The edge gateway receives a large amount of first-piece fabric video captured by the camera and stores it in a distributed industrial database to create several defect detection models.

[0017] The first fabric video in the distributed industrial database is converted into fabric images, and sample data augmentation is performed on each fabric image.

[0018] Each fabric image is labeled with its pattern, defect location, and defect type. Based on the labels, the fabric images are grouped, and each group of fabric images is divided into a training set and a validation set according to a preset ratio.

[0019] Based on the grouping, the corresponding defect detection model is trained using the training set, and the trained defect detection model is validated using the validation set.

[0020] Furthermore, the defect detection model is created based on the Faster R-CNN algorithm, YOLO algorithm, or FastFlow algorithm.

[0021] Furthermore, the distributed industrial database includes a MySQL relational database, a MongoDB unstructured database, an industrial real-time cache database, and an engineering information management database.

[0022] Furthermore, the sample data amplification operation for each fabric image specifically involves:

[0023] Perform sample data augmentation operations on each fabric image, including at least cropping, rotation, scaling, and grayscale conversion.

[0024] Furthermore, during the training process of the defect detection model, the parameter weights of the defect detection model are continuously adjusted through backpropagation using the GIoU loss function to narrow the gap between the predicted bounding box and the true bounding box until the defect detection model converges.

[0025] Furthermore, the specific steps for obtaining defect detection results by using the defect detection model to detect defects in the lace fabric are as follows:

[0026] The edge gateway receives real-time video of the second piece of lace fabric captured by the camera, and matches the pattern in the second piece of fabric video with the corresponding defect detection model.

[0027] The edge gateway inputs the second fabric video into the matching defect detection model to perform defect detection and obtain the defect detection result.

[0028] Furthermore, the second fabric video was used to continue training and optimizing the defect detection model.

[0029] Furthermore, the defect detection results include at least the detection time, pattern, defect location, defect category, defect image, lace fabric information, and fabric production line operating parameters.

[0030] Furthermore, the control of the fabric production line based on cloud platform or host computer control commands specifically involves:

[0031] The edge gateway establishes a communication connection with the fabric production line based on RS485, RS232, Modbus, or Wi-Fi protocols. After receiving control commands from the cloud platform or host computer, it uses middleware to control the fabric production line to adjust its operating frequency and operating status based on the defect detection results. The operating status is forward rotation, reverse rotation, or shutdown.

[0032] The advantages of this invention are:

[0033] An edge gateway is set up to connect to the fabric production line, cameras, cloud platform, and host computer. The edge gateway is used to create and train a defect detection model, which detects defects in the lace fabric. The detection results are then sent to the host computer and cloud platform, which control the fabric production line based on control commands. The fabric production line produces, transports, and displays the lace fabric. Cameras capture video of the fabric production line. The cloud platform and host computer store and display the defect detection results and remotely control the fabric production line. Because the fabric images are grouped according to different patterns, the grouped images are used to train the corresponding defect detection models, combined with sample data augmentation. Increasing the amount of training data effectively improves the defect detection accuracy of the defect detection model for specific patterns of lace fabric. During the training process, the defect detection model continuously adjusts its parameter weights through backpropagation using the GIoU loss function. In actual defect detection, the model is further trained and optimized using second-hand fabric videos, resulting in continuous improvement in detection accuracy. By using a defect detection model located at the edge gateway to perform lace fabric defect detection based on real-time acquired second-hand fabric videos, the computing power of the edge gateway is effectively utilized. Combined with a defect detection model that matches the corresponding pattern, defect detection becomes more targeted, ensuring speed and ultimately greatly improving the accuracy, generalization ability, and timeliness of lace fabric defect detection. Attached Figure Description

[0034] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0035] Figure 1 This is a schematic diagram of the structure of a lace fabric defect detection system based on a cloud platform and edge gateway according to the present invention.

[0036] Figure 2 This is a schematic diagram of the process for detecting defects in lace fabric according to the present invention. Detailed Implementation

[0037] The overall idea of ​​the technical solution in this application embodiment is as follows: Fabric images are grouped according to different patterns, and then the grouped fabric images are used to train the corresponding defect detection models. Combined with sample data augmentation, the amount of training data is effectively increased. During the defect detection model training process, the parameter weights of the defect detection model are continuously adjusted through backpropagation using the GIoU loss function. In the actual defect detection process, the acquired second fabric video is also used to further train and optimize the defect detection model, thereby improving the accuracy and generalization ability of lace fabric defect detection. The defect detection model located at the edge gateway performs lace fabric defect detection based on the real-time acquired second fabric video, effectively utilizing the computing power of the edge gateway. Combined with the defect detection model matching the corresponding pattern, defect detection becomes more targeted, thereby improving the timeliness of lace fabric defect detection.

[0038] Please refer to Figures 1 to 2 As shown, a preferred embodiment of the lace fabric defect detection system based on a cloud platform and an edge gateway of the present invention includes at least one edge gateway, several fabric production lines, several cameras, at least one cloud platform, and at least one host computer; the edge gateway is connected to the fabric production line, cameras, cloud platform, and host computer respectively.

[0039] The edge gateway is used to create and train a defect detection model, and to detect defects in lace fabric through the defect detection model to obtain defect detection results. The defect detection results are then sent to the host computer and cloud platform, and the operation of the fabric production line is controlled based on the control instructions of the cloud platform or the host computer.

[0040] The edge gateway includes a data acquisition module, a local database, a production line operation control module, a defect detection algorithm module, a lace fabric defect repair module, and a data communication interface module.

[0041] The data acquisition module integrates communication drivers for multiple fabric production lines, and is compatible with protocols such as RS485, RS232, Modbus, and Wi-Fi.

[0042] The local database consists of a real-time cache database (responsible for storing real-time data such as the operating speed and status of the fabric inspection machine and other devices), an unstructured database (responsible for storing images, videos and a small amount of label image data), a relational database (responsible for storing and managing lace fabric product information and defect information), and an engineering information database (responsible for storing I / O point information of related equipment in the fabric production line).

[0043] The defect detection algorithm module can automatically select the best defect detection model for different products. The defect detection model is created by supervised algorithms such as Faster R-CNN and YOLO series or unsupervised algorithms such as FastFlow. It can accurately identify the defect location and defect category of lace fabric and mark them. For example, the defect detection model is based on the YOLOv5 convolutional neural network model to locate the fabric defect location in the video frame, give the bounding box of the defect location region, and classify the defect category. In specific implementation, the input fabric image can be preprocessed first, including operations such as cropping, rotating, scaling, and grayscale conversion, to adapt to the input format and size of the YOLOv5 convolutional neural network model and enhance the image contrast and clarity. Then, the preprocessed fabric image is input into the YOLOv5 convolutional neural network model, which outputs the confidence score, coordinates, length and width, and category label of each defect. Based on the output results, each defect is screened and filtered to remove defects with confidence scores below a set threshold or excessive overlap, and classified and statistically analyzed according to the category label.

[0044] The production line operation control module integrates an industrial soft controller (middleware) and equipment control configuration algorithm programs. The industrial soft controller consists of a memory database with a specific data format, an algorithm execution process, and a task scheduling process. It is used to interpret and process the equipment control configuration algorithm programs loaded by the controller. The edge gateway can convert the calculation results of the algorithm logic of the industrial soft controller into control instructions for the fabric production line. The equipment control configuration algorithm programs are driven by the algorithm execution process. After detecting the defect location, the control program automatically controls the running speed and start / stop status of the machines on the fabric production line, thereby adapting to manual re-inspection or sewing operations.

[0045] The data communication interface module integrates standard protocol data interfaces (including HTTP, MQTT, and OPC UA), enabling it to upload data from the edge gateway to the cloud platform, and also to distribute data and models from the cloud platform to the edge gateway.

[0046] The fabric production line is used to produce, convey, and display lace fabric;

[0047] The camera is used to film the fabric production line to collect videos of lace fabric. In some specific scenarios, the camera can also capture and save video images of manual operations during the lace fabric inspection process, including manually marking defective locations and sewing defective locations.

[0048] The cloud platform is used to store and display defect detection results and to remotely control the operation of the fabric production line;

[0049] The cloud platform includes a cloud-based multimodal database, an algorithm model library, and a cloud-based application system. The cloud-based multimodal database functions the same as the local database and is used to synchronize local database information of the edge gateway. The algorithm model library contains algorithm models pre-trained for different patterns, product quality evaluation models, and pattern retrieval and comparison models. In addition, the algorithm model library also includes a model management module, which is responsible for cloud-based training and deployment of models.

[0050] The cloud application system consists of several modules, including a defect detection visualization dashboard, an equipment status monitoring dashboard, a quality evaluation module, and a pattern retrieval and comparison dashboard. The defect detection visualization dashboard provides a more intuitive view of the classification and identification results of different types of defects at the edge layer, as well as real-time video images of the edge environment. The equipment status monitoring dashboard allows real-time observation of the operating status of edge layer mechanical devices and the configuration of various parameters. The quality evaluation module evaluates and provides feedback on the quality grade of the fabric based on the defect detection results. The pattern retrieval module uses a pattern retrieval comparison model (such as an image search algorithm) to quickly find and locate the pattern closest to the target pattern from the pattern database, and then selects pre-trained models of similar patterns as pre-trained models for the target pattern.

[0051] The host computer is used to store and display defect detection results, remotely control the operation of the fabric production line, and push the defect detection results to pre-associated mobile terminals in real time.

[0052] The host computer is deployed at the edge and is responsible for displaying real-time video of collected fabric defects and inspection personnel operations, and issuing alarms for abnormal situations. It can also input control commands and send them to the edge gateway to adjust the real-time operating status of the fabric production line. It mainly includes the following five functional modules: 1. Equipment Start / Stop: Performing basic operations on the equipment, such as starting and stopping mechanical devices and cameras; 2. Real-time Detection Screen: Reading and displaying camera images; 3. Fabric Production Line Control: Controlling parameters such as the operating speed of the fabric production line; 4. Log Information: Displaying log information during the operation process; 5. Other Settings: Additional settings such as automatically saving defect images locally and automatically uploading to the cloud platform's database.

[0053] In practice, edge gateways and cloud platforms can establish bidirectional communication through standard data interfaces such as HTTP, MQTT, and OPC UA to upload and download data and algorithm models.

[0054] When the cloud platform and edge gateway interact with each other, they store unstructured and structured data separately to improve the efficiency of staff, enabling them to locate defects and the batches of defective products more quickly, and to process the batch of products more precisely. This includes calling quality evaluation algorithms to grade the quality of the lace fabric based on statistical data such as the number, type, and confidence level of defects.

[0055] The creation and training of the defect detection model specifically involves:

[0056] The edge gateway receives a large amount of first-piece fabric video captured by several high-definition cameras and stores it in a distributed industrial database to create several defect detection models.

[0057] The first fabric video in the distributed industrial database is converted into fabric images, and sample data augmentation is performed on each fabric image.

[0058] Each fabric image is labeled with its pattern, defect location, and defect type. Based on the labels, the fabric images are grouped, and each group is divided into a training set and a validation set according to a preset ratio. The defect type is at least one of the following: normal, missing stitch, broken stitch, stray thread, extra yarn, broken weft, and broken warp. Before grouping, the fabric images are cut to a specified size according to a preset ratio. The defect locations are also selected using a real bounding box.

[0059] Based on the grouping, the corresponding defect detection model is trained using the training set, and the trained defect detection model is validated using the validation set.

[0060] The defect detection model is created based on the Faster R-CNN algorithm, YOLO algorithm, or FastFlow algorithm.

[0061] The distributed industrial database includes a MySQL relational database, a MongoDB unstructured database, an industrial real-time cache database, and an engineering information management database.

[0062] The specific steps for performing sample data augmentation on each fabric image are as follows:

[0063] Perform sample data augmentation operations on each fabric image, including at least cropping, rotation, scaling, and grayscale conversion.

[0064] During the training process of the defect detection model, the parameter weights of the defect detection model are continuously adjusted through backpropagation using the GIoU loss function to narrow the gap between the predicted bounding box and the ground truth bounding box until the defect detection model converges.

[0065] The training process is as follows: the training set is input into the backbone network of the defect detection model. After multiple convolutions and image slicing, feature layers of different sizes are obtained. These different feature layers are then input into the neck layer. After processing, new feature layers are generated and input to the output. Based on the generated feature layers, the predicted bounding boxes and confidence scores of defects are given. The models are sorted according to their confidence scores. The loss is calculated using the GIoU loss function. The weights are continuously adjusted to narrow the gap between the predicted bounding boxes and the true bounding boxes. Backpropagation is performed using the loss function to adjust the model's weights. The above process is repeated to make the model gradually converge. The parameters are continuously adjusted through testing on the validation set to improve generalization ability and accuracy.

[0066] The specific steps for obtaining defect detection results by using the defect detection model to detect defects in lace fabric are as follows:

[0067] The edge gateway receives real-time video of the second piece of lace fabric captured by the camera, and matches the pattern in the second piece of fabric video with the corresponding defect detection model.

[0068] The edge gateway inputs the second fabric video into the matching defect detection model to perform defect detection and obtain the defect detection result.

[0069] In specific implementation, when no defect detection model corresponding to the current pattern can be matched, the defect detection model of a similar pattern is automatically matched as a pre-trained model through an image comparison search algorithm, and the pre-trained model is trained using the first fabric video and the second fabric video.

[0070] The image comparison search algorithm can use a hash function to map high-dimensional image features to a low-dimensional binary space, or it can use a deep neural network to extract semantic features of the image, or use local image feature descriptors (such as SIFT, SURF, etc.) to extract key points and descriptors of the image, and then use matching algorithms (such as RANSAC, FLANN, etc.) to perform similarity calculation and matching.

[0071] The second fabric video was also used to continue training and optimizing the defect detection model.

[0072] The defect detection results shall include at least the detection time, pattern, defect location, defect category, defect image, lace fabric information, and fabric production line operating parameters.

[0073] The specific operation of the fabric production line controlled by the control commands based on the cloud platform or host computer is as follows:

[0074] The edge gateway establishes a communication connection with the fabric production line based on RS485, RS232, Modbus, or Wi-Fi protocols. After receiving control commands from the cloud platform or host computer, it uses middleware to control the fabric production line to adjust its operating frequency and operating status based on the defect detection results. The operating status is forward rotation, reverse rotation, or shutdown.

[0075] The middleware (industrial soft controller) adopts a standard data structure and can interpret and process the loaded control configuration algorithm. The edge gateway can convert the calculation results of the algorithm logic running in the middleware into control command parameters for the fabric production line, and encapsulate and transmit them to the fabric production line according to the protocol format through the driver.

[0076] Specifically, after detecting a defect, the edge gateway can calculate the distance the defect moves during a shutdown operation based on the preset actual rotation speed of the fabric production line. This moving distance should meet preset constraints, such as the defect falling within the camera's field of view or the defect moving distance not exceeding the maximum movable range of the fabric in the production environment. Additionally, the edge gateway can define a rotation speed control dictionary table to convert the rotation speed value sent to the middleware memory area into numerical values ​​corresponding to the actual rotation speed of the fabric production line. The middleware, based on a preset algorithm calculation cycle (e.g., 50ms) and the received fabric production line rotation speed value, executes the loaded control configuration algorithm and generates control command parameters, which are then sent to the fabric production line to complete the control task. In practice, the data types for reading and writing data in the edge gateway's memory area include AI, AO, LA, DI, DO, LD, etc., supporting integer, character, boolean, and floating-point types. The preset address range for each memory area in the middleware is 0-26999, and the rotation speed value sent to the middleware can be defined between -500 and 4000.

[0077] In practice, service providers can leverage the cloud platform to accumulate fabric defect data, pattern data, detection algorithm models for each pattern, and user-generated labeling data for new defects uploaded by factories. They can then continuously optimize and improve the models: increasing the model's recognition accuracy through the accumulation of more defect data; optimizing the model and improving detection performance through incremental data training; and collecting user feedback on new pattern defects through user-generated labeling to further optimize the model and improve detection performance.

[0078] Service providers can generate revenue by leasing various algorithms on a monthly or yearly basis, including different pattern matching algorithms and fabric quality evaluation algorithms. They can also differentiate pricing by classifying fabrics with different patterns to meet the needs of different customers. For example, when a user needs to detect a new pattern, the cloud platform can recommend detection models with similar patterns to the user. After purchasing, the user can upload their own new pattern fabric data to retrain the algorithm model, thereby optimizing the model's accuracy and quickly iterating to develop a model that can be applied to the new pattern.

[0079] Users can train their own models using incremental data (i.e., data collected by the users themselves) and publish their models and patterns to the cloud platform to sell as a product to others. This can provide users with more business opportunities and make the data on the entire cloud platform more complete and accurate.

[0080] In summary, the advantages of this invention are:

[0081] An edge gateway is set up to connect to the fabric production line, cameras, cloud platform, and host computer. The edge gateway is used to create and train a defect detection model, which detects defects in the lace fabric. The detection results are then sent to the host computer and cloud platform, which control the fabric production line based on control commands. The fabric production line produces, transports, and displays the lace fabric. Cameras capture video of the fabric production line. The cloud platform and host computer store and display the defect detection results and remotely control the fabric production line. Because the fabric images are grouped according to different patterns, the grouped images are used to train the corresponding defect detection models, combined with sample data augmentation. Increasing the amount of training data effectively improves the defect detection accuracy of the defect detection model for specific patterns of lace fabric. During the training process, the defect detection model continuously adjusts its parameter weights through backpropagation using the GIoU loss function. In actual defect detection, the model is further trained and optimized using second-hand fabric videos, resulting in continuous improvement in detection accuracy. By using a defect detection model located at the edge gateway to perform lace fabric defect detection based on real-time acquired second-hand fabric videos, the computing power of the edge gateway is effectively utilized. Combined with a defect detection model that matches the corresponding pattern, defect detection becomes more targeted, ensuring speed and ultimately greatly improving the accuracy, generalization ability, and timeliness of lace fabric defect detection.

[0082] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A defect detection system for lace fabric based on a cloud platform and edge gateway, characterized in that: It includes at least one edge gateway, several fabric production lines, several cameras, at least one cloud platform, and at least one host computer; the edge gateway is connected to the fabric production line, cameras, cloud platform, and host computer respectively; The edge gateway is used to create and train a defect detection model, and to detect defects in lace fabric through the defect detection model to obtain defect detection results. The defect detection results are then sent to the host computer and cloud platform, and the operation of the fabric production line is controlled based on the control instructions of the cloud platform or the host computer. The fabric production line is used to produce, convey, and display lace fabric; The camera is used to film the fabric production line to capture video of lace fabric. The cloud platform is used to store and display defect detection results and to remotely control the operation of the fabric production line; The host computer is used to store and display defect detection results and remotely control the operation of the fabric production line; The creation and training of the defect detection model specifically involves: The edge gateway receives a large amount of first-piece fabric video captured by the camera and stores it in a distributed industrial database to create several defect detection models. The first fabric video in the distributed industrial database is converted into fabric images, and each fabric image is subjected to sample data augmentation operations including at least cropping, rotation, scaling, and grayscale conversion. Each fabric image is labeled with its pattern, defect location, and defect type. Based on the labels, the fabric images are grouped, and each group of fabric images is divided into a training set and a validation set according to a preset ratio. Based on the grouping, the corresponding defect detection model is trained using the training set, and the trained defect detection model is validated using the validation set. The defect detection model is created based on the Faster R-CNN algorithm, YOLO algorithm, or FastFlow algorithm. The distributed industrial database includes a MySQL relational database, a MongoDB unstructured database, an industrial real-time cache database, and an engineering information management database.

2. The lace fabric defect detection system based on a cloud platform and edge gateway as described in claim 1, characterized in that: During the training process of the defect detection model, the parameter weights of the defect detection model are continuously adjusted through backpropagation using the GIoU loss function to narrow the gap between the predicted bounding box and the ground truth bounding box until the defect detection model converges.

3. The lace fabric defect detection system based on a cloud platform and edge gateway as described in claim 1, characterized in that: The specific steps for obtaining defect detection results by using the defect detection model to detect defects in lace fabric are as follows: The edge gateway receives real-time video of the second piece of lace fabric captured by the camera, and matches the pattern in the second piece of fabric video with the corresponding defect detection model. The edge gateway inputs the second fabric video into the matching defect detection model to perform defect detection and obtain the defect detection result.

4. The lace fabric defect detection system based on a cloud platform and edge gateway as described in claim 3, characterized in that: The second fabric video was also used to continue training and optimizing the defect detection model.

5. The lace fabric defect detection system based on a cloud platform and edge gateway as described in claim 1, characterized in that: The defect detection results shall include at least the detection time, pattern, defect location, defect category, defect image, lace fabric information, and fabric production line operating parameters.

6. The lace fabric defect detection system based on a cloud platform and edge gateway as described in claim 1, characterized in that: The specific operation of the fabric production line controlled by the control commands based on the cloud platform or host computer is as follows: The edge gateway establishes a communication connection with the fabric production line based on RS485, RS232, Modbus, or Wi-Fi protocols. After receiving control commands from the cloud platform or host computer, it uses middleware to control the fabric production line to adjust its operating frequency and operating status based on the defect detection results. The operating status is forward rotation, reverse rotation, or shutdown.

Images