Fabric defect detection device and fabric defect detection method

Abstract

The invention provides a fabric defect detection device and a fabric defect detection method, and belongs to the field of fabric detection.The fabric defect detection method specifically comprises the steps that a camera shooting unit continuously shoots fabric in real time to obtain real-time images; the recognition unit analyzes the real-time image and determines the area where defects on the fabric are located; the control unit adjusts the conveying speed of the conveying unit according to the recognized area where the defects are located so as to renovate the defects. And the recognition unit inputs a shot real-time image into a pre-trained deep learning neural network model to obtain an output result of the model, and performs non-maximum suppression and threshold filtering on the output result to obtain a final defect type on the fabric and a corresponding image coordinate, so as to determine the position of the defect on the fabric. According to the processing scheme, cloth inspecting workers are replaced to accurately detect and position the fabric defects, the cloth inspecting workers are concentrated on cloth cover trimming, and therefore the cloth inspecting efficiency and the cloth inspecting quality are greatly improved, and meanwhile the cloth inspecting work intensity is reduced.

Description

Technical Field

[0001] The present invention relates to the field of fabric detection, and in particular to a fabric defect detection device and a fabric defect detection method. Background Art

[0002] Due to problems such as raw materials and production processes, various defects will occur in the fabric during the weaving process. These defects need to be detected after the fabric is woven. The fabric is spread out through a cloth roller and then passed through a light box. The light box illuminates the fabric and the fabric inspector uses his eyes to detect the defects on the fabric. However, the manual inspection method is inefficient and has a low detection rate. Different fabric inspectors vary greatly and cannot accurately grade the fabric.

[0003] Therefore, a device is needed to replace manual labor and automatically complete the defect detection work after fabric weaving. Summary of the invention

[0004] Therefore, in order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a fabric defect detection device and a fabric defect detection method for accurately identifying fabric defects.

[0005] In order to achieve the above-mentioned purpose, the present invention provides a fabric defect detection device, comprising: a camera unit, used for real-time continuous shooting of fabric to obtain real-time images; an identification unit, used for analyzing the real-time images and determining the locations of defects on the fabric; a control unit, used for adjusting the conveying speed of the conveying unit according to the identified locations of the defects, and controlling the fabric to stop the area where the defects are located on a cloth repairing light box to facilitate the repair of defects, wherein the identification unit inputs the real-time image to be analyzed into a pre-trained deep learning neural network model and performs non-maximum suppression and threshold filtering to obtain the output result of the model, wherein the output result includes the detected defect type and image coordinates, thereby determining the location of the defects on the fabric.

[0006] In one embodiment, the recognition unit further extracts a plurality of continuous real-time images within a predetermined period, and recognizes fabric boundaries in the real-time images; the recognition unit further determines a fabric boundary range based on the recognized fabric boundaries.

[0007] In one embodiment, the training method of the deep learning model includes: obtaining a sample data set composed of sample images, wherein one sample image is associated with a defect label group; the defect label group includes at least one defect label type and at least one defect label area; performing image enhancement operations on the image before training to improve the generalization ability of the model; the image enhancement operations include at least one of random rotation of the image, random brightness, contrast, and saturation transformation of the image; inputting the sample data into a neural network model for forward propagation, and obtaining the loss value of the model through a designed loss function; performing backpropagation on the loss value in the neural network model to obtain the gradient value corresponding to each parameter in the model; using the gradient value to iteratively update the model parameters of the neural network model; repeating the model forward propagation and backpropagation and parameter updating process until the change in the loss value is within a preset threshold, determining that the model training is completed, saving the model parameters at this time, and obtaining the final neural network model.

[0008] In one embodiment, the method further includes performing an image normalization operation on the sample images in the sample data set.

[0009] In one embodiment, the camera unit includes a camera and a shooting light box. The shooting light box is located below the moving path of the fabric to provide fill-in light for the fabric. The camera is located above the fabric to perform real-time continuous shooting of the fabric filled with light by the shooting light box.

[0010] In one embodiment, it also includes a cloth repairing device for repairing the defects to achieve a "pre-defect then repair" mode.

[0011] A fabric defect detection method is applied in the above-mentioned fabric defect detection device, comprising: taking real-time continuous photos of the fabric to obtain a real-time image; inputting the real-time image into a pre-trained deep learning neural network model and performing non-maximum suppression and threshold filtering to obtain an output result of the model, wherein the output result includes the detected defect type and image coordinates, thereby determining the location of the defect on the fabric; adjusting the conveying speed of the conveying unit according to the identified location of the defect, and controlling the fabric to stop the defect area on a cloth repairing light box to facilitate the repair of the defect.

[0012] Compared with the prior art, the advantages of the present invention are: real-time continuous shooting can avoid misjudgment or omission of defects, and by analyzing real-time images through a deep learning network model, defects can be quickly identified and the grade of the fabric can be determined; defects can also be displayed according to their locations for easy repair. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0014] Figure 1 is a schematic structural diagram of a fabric defect detection device in an embodiment of the present invention;

[0015] Figure 2 It is a flow chart of the training method of the deep learning network model in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0017] The following describes the implementation methods of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The present application can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, in the absence of conflict, the following embodiments and the features in the embodiments can be combined with each other. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work belong to the scope of protection of the present application.

[0018] It should be noted that various aspects of the embodiments within the scope of protection of the present invention are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present application, it should be understood by those skilled in the art that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number and aspect described herein can be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this device and / or practice this method.

[0019] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present application. The drawings only show components related to the present application rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0020] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0021] like Figure 1 As shown, an embodiment of the present application provides a fabric defect detection device, including a camera unit, a recognition unit and a control unit.

[0022] The camera unit is used to continuously shoot the fabric in real time to obtain a real-time image. In one embodiment, the camera unit may include only one camera, or may include multiple cameras and a light box for fill light; but each camera can completely shoot the fabric in the area perpendicular to the fabric loading direction. In one embodiment, Figure 1 As shown, the camera unit includes a camera 2 and a shooting light box 1. The shooting light box 1 is located below the fabric moving path and provides fill light for the fabric. The camera 2 is located above the fabric and performs real-time continuous shooting of the fabric that is filled light by the shooting light box 1.

[0023] The recognition unit is used to analyze the real-time image and determine the location of the defect on the fabric. The recognition unit can receive the real-time image sent by the camera unit through wireless transmission, or through wired transmission. The recognition unit can be various chips with computing functions, such as FPGA chips, GPU chips, etc.; it can also be a smart device, such as a computer, a mobile phone, a server, etc.

[0024] The recognition unit inputs the real-time image to be analyzed into a pre-trained deep learning neural network model and performs non-maximum suppression and threshold filtering to obtain the output result of the model, which includes the detected defect type and image coordinates, and then determines the location of the defect on the fabric.

[0025] There is data communication between the control unit and the identification unit. The control unit is used to adjust the conveying speed of the conveying unit according to the identified defect area, and control the fabric to stop the defect area on the cloth repairing light box 4 to facilitate subsequent repair of the defect. The conveying unit can be composed of a cloth feeding roller 3 and a cloth receiving roller 5. In one embodiment, the control unit and the identification unit can be set on the same chip or device, or on different chips or devices.

[0026] In one embodiment, the fabric defect detection method comprises the following steps:

[0027] Take real-time and continuous photos of fabrics to obtain real-time images;

[0028] The real-time image to be analyzed is input into the pre-trained deep learning neural network model and non-maximum suppression and threshold filtering are performed to obtain the output result of the model. The output result includes the detected defect type and image coordinates, and then the location of the defect on the fabric is determined;

[0029] The conveying speed of the conveying unit is adjusted according to the identified defect area, and the fabric is controlled to stop the defect area on the fabric repairing light box to facilitate subsequent defect repair.

[0030] The above-mentioned device and method can avoid misjudgment or omission of defects through real-time continuous shooting, and can quickly identify defects and determine the grade of fabrics by analyzing real-time images through a deep learning neural network model; defects can also be displayed according to the area where the identified defects are located.

[0031] In one embodiment, the recognition unit further extracts a plurality of continuous real-time images within a predetermined period, and recognizes fabric boundaries in the real-time images; the recognition unit further determines a boundary range of the fabric based on the recognized fabric boundaries.

[0032] The recognition unit can extract multiple real-time images continuously within a predetermined period and recognize the fabric boundary in the real-time image. Specifically, the recognition unit can superimpose multiple real-time images, and simultaneously determine the fabric-up direction and the fabric boundary of the fabric through a changing dynamic layout; the recognition unit can also determine the fabric boundary based on the fabric-up direction; the recognition unit can also select an area from the real-time image, and then determine the fabric-up direction of the fabric according to the movement of the area within a predetermined period, and then determine the fabric boundary.

[0033] like Figure 2 As shown, in one embodiment, the training method of the deep learning neural network model includes:

[0034] Step 201 : acquiring a sample data set consisting of sample images, wherein one sample image is associated with one defect label group; the defect label group includes at least one defect label type and at least one defect label area.

[0035] The server can obtain a sample data set composed of sample images, and one sample image is associated with a defect label group; the defect label group includes at least one defect label type and at least one defect label area. Among them, defects are various defects generated on the fabric surface during the weaving process due to the influence of factors such as raw materials, semi-finished products, production equipment and operation. Defect label types can be divided into "warp defects", "weft defects", "intensive defects", "damage defects", "single defects", "oil stain defects" and so on. Fabric defects are an important factor affecting fabric quality and an important indicator of the production and management level of an enterprise. A sample image can contain at least one defect or multiple defects. The resolution of the sample images in the sample data set can be variable or consistent. The defect label area includes the position data and size data of the defect in the sample image. In one embodiment, the defect label area is the coordinate area range of the defect that can be covered by a rectangular frame with a minimum area.

[0036] Step 202: perform image enhancement on the sample images in the sample data set. The image enhancement operation includes at least one of random rotation of the image and random transformation of the brightness, contrast, and saturation of the image.

[0037] The server can perform image enhancement on the sample images in the sample data set to obtain enhanced sample data to improve the generalization ability of the model. The angle of rotation in the random rotation of the image changes within a small range, which can simulate the state of the actual fabric during the detection process. The random changes in brightness, contrast, and saturation can model the different lighting states of the fabric during the actual detection process and improve the generalization ability of the model.

[0038] Step 203: input the sample data into the deep learning neural network model for forward propagation of the model, and calculate the loss value of the sample data through the designed loss function.

[0039] The server inputs the sample data into the deep learning neural network model for forward propagation of the model, and calculates the loss value of the sample data through the designed loss function.

[0040] Step 204, back-propagates the loss value in the model, calculates the gradient value corresponding to each parameter in the model, and uses the gradient value to iteratively update the model parameters.

[0041] The server back-propagates the loss value in the model, calculates the gradient value corresponding to each parameter in the model, and uses the gradient value to iteratively update the model parameters.

[0042] Step 205, repeat the operations of forward propagation, back propagation and parameter update in the model for all sample data until the loss value of the model is determined to be within a preset threshold. At this time, the parameter value of the model is saved to obtain the final deep learning neural network model.

[0043] The server repeats the operations of forward propagation, back propagation, and parameter update in the model for all sample data until it determines that the loss value of the model is within the preset threshold (including the threshold no longer decreasing). At this time, the parameter value of the model is saved to obtain the final deep learning neural network model.

[0044] In one embodiment, a normalization operation is performed on sample images in the sample data set.

[0045] In one embodiment, a cloth repairing light box 4 is also included.

[0046] The cloth repairing light box 4 is located between the camera 2 and the cloth collecting roller 5, below the cloth, and is used to fill in the fabric with light in an inclined manner upward so as to repair the defects. The angle of the light inclination can be adjusted as needed.

[0047] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed in the present application should be included in the protection scope of the present application.

Claims

1. A fabric defect detection device, characterized in that: include: The camera unit is used to continuously shoot the fabric in real time to obtain real-time images; An identification unit, used for analyzing the real-time image and determining the location of the defect on the fabric; A control unit is used to adjust the conveying speed of the conveying unit according to the identified location of the defect, and control the fabric to stop the area where the defect is located on the fabric repairing light box to facilitate repairing the defect. Among them, the recognition unit inputs the real-time image to be analyzed into a pre-trained deep learning neural network model and performs non-maximum suppression and threshold filtering to obtain the output result of the model. The output result includes the detected defect type and image coordinates, thereby determining the location of the defect on the fabric.

2. The fabric defect detection device according to claim 1, characterized in that: The recognition unit further extracts a plurality of continuous real-time images within a predetermined period and recognizes fabric boundaries in the real-time images; the recognition unit further determines a fabric boundary range based on the recognized fabric boundaries.

3. The fabric defect detection device according to claim 1, characterized in that: The training method of the deep learning neural network model includes: Acquire a sample data set consisting of sample images, wherein one sample image is associated with one defect label group; the defect label group includes at least one defect label type and at least one defect label area; Performing image enhancement on sample images in the sample data set, where the image enhancement operation includes at least one of random rotation of the image, random brightness, contrast, and saturation transformation of the image; Input the sample data into the neural network model for forward propagation, and obtain the loss value of the model through the designed loss function; Back-propagating the loss value in the neural network model to obtain a gradient value corresponding to each parameter in the model, and using the gradient value to iteratively update the model parameters of the neural network model; Repeat the process of model forward propagation, back propagation and parameter updating until the change in loss value is within the preset threshold. The model training is considered complete, the model parameters at this time are saved, and the final neural network model is obtained.

4. The fabric defect detection device according to claim 3, characterized in that: The method also includes performing an image normalization operation on the sample images in the sample data set.

5. The fabric defect detection device according to claim 1, characterized in that: The camera unit includes a camera and a shooting light box. The shooting light box is located below the moving path of the fabric to provide fill light for the fabric; The camera is located above the fabric and performs real-time continuous shooting of the fabric supplemented by the shooting light box.

6. The fabric defect detection device according to claim 5, characterized in that: It also includes a cloth repairing unit for repairing the defects.

7. A fabric defect detection method, characterized in that: The fabric defect detection device used in any one of claims 1 to 6 comprises: Take real-time and continuous photos of fabrics to obtain real-time images; The real-time image is input into a pre-trained deep learning neural network model and non-maximum suppression and threshold filtering are performed to obtain the output result of the model, wherein the output result includes the detected defect type and image coordinates, thereby determining the location of the defect on the fabric; The conveying speed of the conveying unit is adjusted according to the identified position of the defect, and the fabric is controlled to stop the area where the defect is located on the fabric repairing light box to facilitate the repair of the defect.

Images