CN112215824A - YOLO-v 3-based cloth cover defect detection and auxiliary device and method - Google Patents
YOLO-v 3-based cloth cover defect detection and auxiliary device and method Download PDFInfo
- Publication number
- CN112215824A CN112215824A CN202011111554.0A CN202011111554A CN112215824A CN 112215824 A CN112215824 A CN 112215824A CN 202011111554 A CN202011111554 A CN 202011111554A CN 112215824 A CN112215824 A CN 112215824A
- Authority
- CN
- China
- Prior art keywords
- cloth
- defect
- defects
- detecting
- yolo
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000007547 defect Effects 0.000 title claims abstract description 141
- 239000004744 fabric Substances 0.000 title claims abstract description 111
- 238000001514 detection method Methods 0.000 title claims abstract description 33
- 238000000034 method Methods 0.000 title claims abstract description 16
- 230000008439 repair process Effects 0.000 claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 14
- 238000007689 inspection Methods 0.000 claims description 21
- 238000010586 diagram Methods 0.000 claims description 15
- 238000003672 processing method Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 7
- 238000007477 logistic regression Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000003064 k means clustering Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 2
- 238000012544 monitoring process Methods 0.000 claims description 2
- 238000012800 visualization Methods 0.000 abstract description 2
- 238000009941 weaving Methods 0.000 description 7
- 238000013135 deep learning Methods 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 238000007730 finishing process Methods 0.000 description 3
- 238000007781 pre-processing Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 208000028804 PERCHING syndrome Diseases 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000004043 dyeing Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000007639 printing Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000004753 textile Substances 0.000 description 1
- 238000004804 winding Methods 0.000 description 1
- 239000002759 woven fabric Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/89—Investigating the presence of flaws or contamination in moving material, e.g. running paper or textiles
- G01N21/8914—Investigating the presence of flaws or contamination in moving material, e.g. running paper or textiles characterised by the material examined
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/04—Programme control other than numerical control, i.e. in sequence controllers or logic controllers
- G05B19/05—Programmable logic controllers, e.g. simulating logic interconnections of signals according to ladder diagrams or function charts
- G05B19/054—Input/output
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30124—Fabrics; Textile; Paper
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Textile Engineering (AREA)
- Quality & Reliability (AREA)
- Automation & Control Theory (AREA)
- Probability & Statistics with Applications (AREA)
- Health & Medical Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Treatment Of Fiber Materials (AREA)
Abstract
The invention discloses a YOLO-v 3-based cloth cover defect detection and auxiliary device and method, which comprises cloth which is produced and operated on line, wherein the cloth is conveyed by a cloth roller, the cloth roller is driven by a variable frequency motor, and the device also comprises an industrial camera, an image acquisition card, a processor, a register, a PLC (programmable logic controller), a display screen, a brake block and an unwinding roller; the brake blocks are respectively arranged on the cloth roller and the unwinding roller; the cloth defect detection is carried out according to the following steps: making a training data set; training the model, storing the weight parameters and implementing automatic detection; writing the detected defect information. The cloth repair assistance can be divided into the following steps: reading the stored defect information; the repair auxiliary device unwinds the cloth to the position of the defect; and the defect information is displayed on a display screen to assist in manual repair. The invention has accurate algorithm detection; the detection speed is high; various defects can be detected simultaneously; better visualization performance, greatly reduced workman's burden.
Description
Technical Field
The invention relates to the technical field of defect detection, in particular to a YOLO-v 3-based cloth cover defect detection and auxiliary device and method.
Background
The cloth inspecting is a weaving after-finishing process, and no matter the cloth is woven cloth or knitted cloth, the link of cloth inspecting is omitted. The cloth inspection is to inspect the defects on the cloth, mark the positions and the types of the defects, count the types and the number of the defects and grade the cloth. Depending on the type of defect, some serious defects may be repaired to meet customer requirements and save costs. The finishing process is the last process before the cloth leaves the factory, and the cloth inspection is carried out on the cloth for weaving, dyeing, printing and the like, so the cloth inspection is a key process in the finishing process.
At present, textile enterprises still widely adopt a manual cloth inspection mode, namely, the quality of the cloth surface is observed by naked eyes, and the marking of the defects of the cloth surface is finished. The cloth is taken off from the weaving machine in a roll form, and after a roll is tested, the cloth is delivered to a special person for repairing. The manual repair mode has the problems of high labor intensity, low efficiency, high omission factor, high labor cost and the like.
The automatic cloth inspecting by the machine is a new mode, and has great application value and market prospect. The core of the automatic perching is the design of an automatic perching algorithm. The method has the main idea that image preprocessing is firstly carried out, such as image acquisition and image segmentation, binarization, gray-scale image conversion and other methods. And then extracting or highlighting the characteristics of the cloth defects, wherein common algorithms comprise Gaussian filtering, wavelet transformation and fast Fourier transformation. And finally, identifying and classifying defects in the image, wherein common algorithms comprise histogram identification, threshold comparison identification, contour transformation identification algorithm and the like. Woven fabrics are very wide, and in order to reduce the complexity of an algorithm and improve the detection speed, multi-step pretreatment is often needed. However, the preprocessing also filters out many characteristics of the defects, so the image detection method has the problems of low identification precision, difficulty in meeting high-speed detection and the like.
In recent years, deep learning has been greatly developed, some deep learning algorithms are beginning to be applied to image detection, the most common is Convolutional Neural Network (CNN), but as a basic deep learning algorithm, the structure is simpler, when a plurality of different defects appear in a cloth cover, the accuracy is lower, the misjudgment rate is higher, and the occurrence range of the defects cannot be given.
Nowadays, high-speed weaving machines are developed for years, the speed of a water-jet weaving machine can reach 1000r/min, the speed of an air-jet weaving machine can also reach 500-.
Disclosure of Invention
The invention aims to provide a YOLO-v 3-based cloth cover defect detection and auxiliary device and method, in particular to a novel automatic defect detection and repair auxiliary device which has high detection accuracy and high detection speed and can meet the requirements of high-speed mass production.
In order to achieve the purpose, the invention adopts the technical scheme that: a cloth cover defect detecting and assisting device based on YOLO-v3 comprises cloth which is produced and operated on line, the cloth is conveyed by a cloth roller, the cloth roller is driven by a variable frequency motor, and the cloth cover defect detecting and assisting device is characterized in that: the device also comprises an industrial camera, an image acquisition card, a processor, a register, a PLC (programmable logic controller), a display screen, a brake block and an unwinding roller; the industrial camera is positioned above the cloth; the unwinding roller is also driven by a variable frequency motor; the brake blocks are respectively arranged on the cloth roller and the unwinding roller; the industrial camera, the image acquisition card, the processor, the register and the PLC are electrically connected, and the PLC is respectively electrically connected with the display screen and the variable frequency motor.
The processing method of the cloth cover defect detection and auxiliary device based on YOLO-v3 is characterized by comprising defect inspection and cloth repair assistance, and comprises the following specific steps:
the defect inspection comprises the following steps:
a. making a training data set and a testing data set;
b. determining the structure and training parameters of the model according to the types of the common defects, importing training data, and starting to train the weight of the model;
c. the weights of the trained models are saved and tested. The automatic monitoring of the cloth defects is realized;
d. when the defect detection device detects defects, writing the types and the position information of the defects into a system;
cloth repair assist
The cloth repair assistance comprises the following steps:
a. reading the stored defect information;
b. rewinding the cloth detected by the defect detection device to the position of the defect;
c. and the defect information is displayed on a display screen to assist in manual repair.
Further, in the defect inspection step a, a defect picture shot by an industrial camera is down-sampled and compressed to 416x416 pixels by using a gaussian pyramid, the compressed defect picture is stored in a commonly used jpeg, png and bmp format, the defect type and position are determined through manual screening, and the defect type and position are written into a txt or xml format file, specifically, a picture name, a defect type and a defect coordinate.
Further, in the defect inspection step b, the input of the model is 416x416x3 three-dimensional RGB preprocessed pictures; outputting feature maps with three different sizes by a defect detection algorithm, and determining the sizes of the feature maps to be 13x13x255, 26x26x255 and 52x52x255 respectively according to the characteristics of the cloth cover defects; defects with different sizes and forms are met, fine objects can be detected by the small-scale characteristic diagram, and coarse-grained objects can be detected by the large-scale characteristic diagram; so that large, medium and small defects can be detected; after defect inspection is completed, the occurrence range of defects needs to be predicted, specifically: dividing the characteristic diagram into a grid, wherein the grid can be refined into a grid unit, each characteristic diagram corresponds to three anchor frames, characteristic diagrams with three scales are set, 9 anchor frames are provided in total, and the sizes of the 9 anchor frames are determined by a k-means clustering method according to the positions of the defects marked in advance; each pixel grid can be subdivided into grid units; the prediction box has four parameters, respectively bx、by、bwAnd bh(ii) a The calculation formula is as follows (1-4):
bx＝σ(tx)+Cx (1)
by＝σ(ty)+Cy (2)
wherein t isx，tyRepresenting the predicted coordinate offset value, tw，thIs scaling; cx，CyIs the coordinates of the grid cells at the upper left corner of the feature map, and the size of each grid cell is 1x 1; p is a radical ofw，phMapping a preset anchor frame to the width and the height of the feature diagram, wherein sigma (·) represents logistic regression, and finely adjusting the prediction frame through the logistic regression;
determining the accuracy degree of the prediction box by using the IOU intersection ratio, wherein the formula is as follows (5):
where A is the prediction box and B is the true box;
determining that each grid unit has 3 kinds of anchor frames, taking the anchor frame with the largest superposition with the real frame as a prediction frame, not participating in prediction by other anchor frames, and then translating and scaling the prediction frame; the parameters of the four prediction frames are optimized by continuously training, learning and adjusting the weight, so that the intersection ratio of the prediction frames and the real frames can reach the maximum;
setting a threshold, when the intersection ratio is smaller than the threshold, the prediction is considered to be wrong and not displayed, and only when the intersection ratio is larger than the threshold, a prediction frame is displayed, wherein the threshold is set to be 0.6;
the loss function is the superposition of four parts, namely loss is made on the predicted central coordinate, loss is made on the width and height of the predicted boundary box, loss is made on the predicted category and loss is made on the predicted confidence coefficient.
Further, in the defect checking step c, the performance of testing the error detection of the model specifically includes: detecting the type and the predicted position of the cloth cover defects, and detecting the time length of a single picture.
Further, in the defect checking step c, the performance of testing the error detection of the model specifically includes: detecting the type and the predicted position of the cloth cover defects, and detecting the time length of a single picture.
Further, the reading of the stored defect information in the cloth repair assisting step a is to read the defect information and the position detected by the defect inspection device.
Further, in the cloth repair assisting step b, the detected cloth is rewound to the corresponding defect position, and the defect type is displayed on the display screen, so as to assist manual repair.
After adopting the structure, the invention has the beneficial effects that:
1) the recognition speed is high, and after the YOLO-v3 algorithm is adopted, only 0.04 second is needed for recognizing one frame of picture, which means that 25 frames of pictures can be detected per second. Compared with other algorithms, the detection time is only one percent of that of FastR-cnn, the identification speed is high, and the method can be completely used for real-time detection.
2) The identification accuracy is high, yarn defects of different sizes and shapes can be identified simultaneously, the accuracy is higher than that of a DSSD algorithm, but the detection time is shorter. The traditional image processing technology needs to perform multi-step preprocessing on the image to highlight the outline of the defect, and then classify the extracted features. Many original characteristics of the defects can be lost by the method, and different defects extracted at the same time can interfere with each other, so that the identification accuracy is reduced.
3) The device has better visualization capacity, different defects are marked by different colors and are visually displayed on the display screen, the defects do not need to be found manually, workers can repair the yarn defect cloth cover conveniently, and the labor amount of the workers is reduced.
4) In summary, the present invention is directed to a novel yarn defect detecting and repairing auxiliary device; the method is realized by using a YOLO-v3 algorithm with high recognition speed and high recognition accuracy, and the method can effectively detect the cloth defects and assist in manual repair; only need be equipped with a high-speed detection machine in actual production, many repair are supplementary just can satisfy the production requirement, and artifical mode divide into two steps: the defect inspection and defect repair need a large amount of manpower and material resources to meet the production requirements.
Drawings
FIG. 1 is a schematic structural view of the present invention;
FIG. 2 is a network structure of YOLO-v3 according to the present invention;
FIG. 3 is a Darknet-53 network architecture according to the present invention;
FIG. 4 is a diagram illustrating a prediction box structure according to the present invention;
FIG. 5 is a diagram illustrating the test results of the present invention.
Description of reference numerals:
1 piece of cloth, 2 cloth rollers, 3 variable frequency motors, 4 industrial cameras, 5 image acquisition cards, 6 processors, 7 registers, 8 PLC controllers, 9 display screens, 10 brake blocks and 11 unwinding rollers.
Detailed Description
The invention will be further described with reference to the accompanying drawings.
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and the detailed description. It should be understood that the detailed description and specific examples, while indicating the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.
Referring to fig. 1, a cloth cover defect detecting and assisting device based on YOLO-v3 comprises a cloth 1 which is produced and operated on line, wherein the cloth is conveyed by a cloth roller 2, the cloth roller 1 is driven by a variable frequency motor 3, and the cloth cover defect detecting and assisting device further comprises an industrial camera 4, an image acquisition card 5, a processor 6, a register 7, a PLC (programmable logic controller) 8, a display screen 9, a brake block 10 and an unwinding roller 11; the industrial camera 4 is positioned above the cloth 1; the unwinding roller 11 is also driven by a variable frequency motor 3; the brake blocks 10 are respectively arranged on the cloth roller 2 and the unwinding roller 11; the industrial camera 4, the image acquisition card 5, the processor 6, the register 7 and the PLC controller 8 are electrically connected, and the PLC controller 8 is respectively electrically connected with the display screen 9 and the variable frequency motor 3; the invention is divided into two parts: a defect inspection section and a cloth repair assist section; the upper part is used for defect inspection, and the lower part is used for cloth repair assistance; the dashed lines represent transmission cables that connect the various components of the device in series; the PLC controller 8 controls the variable frequency motor 3.
Inspection of defects
The cloth is unwound from the weaving shaft, the industrial camera 4 (preferably an industrial COD camera) scans the cloth, the image acquisition card 5 acquires each frame of picture transmitted by the industrial camera 4 and inputs the frame of picture into the processor 6, and the YOLO-v3 algorithm inspects each frame of picture, records the defect type and the coordinate position and stores the defect type and the coordinate position in the register 7. The cloth roller winds the cloth at a certain speed.
Training data set preparation
And collecting pictures containing the defects shot by the industrial camera 4, manually marking the image names, the defect types and the defect coordinate positions. In order to avoid the influence of local too bright or too dark of an image on the identification accuracy rate caused by uneven illumination of a workshop, brightness normalization needs to be firstly carried out on an original image. The width of the loom is large, the size of the whole shot picture is large, the requirement on deep learning equipment is high, and network training and identification are not facilitated, so that two times of Gaussian pyramid down-sampling are sequentially performed on the image, and each frame of picture is controlled to be 416x416 pixels. The picture size is reduced to 416x416 pixels, and the marked coordinate frame is also reduced according to the corresponding proportion.
Model structure
The influence of the recognition speed and the accuracy is comprehensively considered, and the novel YOLO-v3 algorithm is considered to be adopted as a defect detection and recognition algorithm.
In order to detect fine defects, shorten the time and improve the detection efficiency, the defect detection part of the algorithm comprises 72 layers of batch normalization processing (Batchnormalization), 75 layers of convolution layers (Conv) and 72 layers of activation function layers, wherein the activation function layers are selected from a leaky linear rectifying unit (Leakyrelu), a 2-layer upper sampling layer (UpSampling) and a 5-layer zero padding layer (Zeropadding). The model has no fully connected layers and the tensor size is varied by the step size of the convolution kernel.
Referring to fig. 2, the input is a three-dimensional RGB pre-processed picture of 416x416x 3. The DBL layer is a basic component of YOLO-v3 and consists of a convolution layer, a batch standardization processing layer and a leakage linear rectifying unit (Leakyrelu). Concat stands for tensor spliced layer, splicing the output of the DBL middle layer with the following upsampled layer. And outputting feature maps with three different sizes. Outputting feature maps with three different sizes by a defect detection algorithm, and determining the sizes of the feature maps to be 13x13x255, 26x26x255 and 52x52x255 respectively according to the characteristics of the cloth cover defects; the latter two feature maps are obtained by up-sampling and twice-combining the former feature map, so as to obtain fine-grained information from the lower-layer features. The three feature maps are used for detecting objects with different sizes, the smaller feature map can detect the finer object, and the larger feature map can detect the coarse-grained object. By the matching, large, medium and small defects can be detected.
Referring to fig. 3, the network is a feature extractor in the yolo-v3 algorithm. More implicit information is extracted by deepening the number of layers. The residual layer (Resnet) is a layer in which outputs (x and y) of two convolutional layers are merged and input to the next layer. The 3 layers in the first block were cycled 1 time, the second block 2 times, the third 8 times, the fourth 8 times, and the fifth 4 times. The sizes of convolution kernels are respectively 3x3 and 1x1, and the number of the convolution kernels is 32, 64, 128, 256, 512 and 1024.
Predictive box algorithm
Referring to fig. 4, the feature map is divided into a grid, the grid can be refined into a grid unit, each feature map corresponds to three anchor frames, there are three feature maps, and 9 anchor frames are determined in total by using a k-means clustering method according to the positions of the defects marked in advance. It is known that the more the number of clusters, the higher the intersection ratio, but the higher the complexity of the corresponding algorithm. Here the number of clusters is 9, i.e. 9 anchor frames are generated.
Each grid may in turn be divided into a plurality of grid cells. The prediction box has four parameters, respectively bx，by，bwAnd bh。
The calculation formula is shown as (1-4):
bx＝σ(tx)+Cx (1)
by＝σ(ty)+Cy (2)
wherein t isx，tyRepresenting the predicted coordinate offset value, tw，thIs scaling; cx，CyIs the coordinates of the grid cells at the upper left corner of the feature map, and the size of each grid cell is 1x 1; p is a radical ofw，phThe preset anchor frame is mapped to the width and the height of the feature diagram, sigma (·) represents logistic regression, and the prediction frame is finely adjusted through the logistic regression.
tx，ty，tw，thThe calculation formula (5) is as follows:
tx＝Gx-Cx (5)
ty＝Gy-Cy (6)
tw＝log(Gw/Pw) (7)
th＝log(Gh/Ph) (8)
wherein G isx，Gy，Gw，GhIs the coordinates of the true prediction box. In training, when the central point of the true prediction box falls into a certain grid cell, 3 anchor boxes of the grid cell are responsible for prediction.
In practice, the degree of accuracy of the prediction box is determined by the IOU (cross-over ratio), and the formula is as follows (9):
where A is the prediction box and B is the true box.
And 3 kinds of anchor frames exist in each grid unit, the anchor frame which is the largest in superposition with the real frame is determined as a prediction frame, the other anchor frames do not participate in prediction, and then the prediction frame is translated and scaled. And optimizing the parameters of the four prediction frames by continuously training and learning to adjust the weights, so that the intersection ratio of the prediction frames and the real frames can be maximized.
And setting a threshold, when the intersection ratio is smaller than the threshold, considering that the prediction is wrong and not displayed, and only when the intersection ratio is larger than the threshold, displaying a prediction frame. Here the threshold is set to 0.6.
The loss function is the superposition of four parts, namely, the predicted central coordinate, the width and the height of a predicted boundary box, the predicted category and the confidence coefficient of the prediction are combined to be used as the loss function
Cloth repair assist
The cloth with the inspected defects is wound to wait for manual repair, the defects are marked by the conventional manual defect inspection, and repair personnel need to inspect the positions of the defects for the second time, so that time and labor are wasted. The repair auxiliary device is designed, and the PLC reads the defect information in the register and respectively controls the two variable frequency motors to rotate in the same direction. The two motors respectively drive the unwinding roller and the cloth roller, the function of the two motors is to unwind the wound cloth and wind the repaired cloth, the rotating speeds of the two motors are kept constant, and the cloth cover is pulled due to the fact that the cloth winding motor is too fast, so that the cloth is accidentally drafted; too slow cloth rolling motor can lead to cloth too loose, be unfavorable for artifical restoration. When the defect coordinates are read, the variable frequency motor starts to reduce the speed, and the brake block starts to gradually hold the unwinding roller and the cloth roller, so that the accidental drafting of the cloth is reduced to the minimum. The cloth is accurately stopped in the area to be repaired, the type and the position of the defects are visually displayed in a display screen, and after the defects are repaired, the machine is restarted to enter the next defect repair.
The above description is only for the purpose of illustrating the technical solutions of the present invention and not for the purpose of limiting the same, and other modifications or equivalent substitutions made by those skilled in the art to the technical solutions of the present invention should be covered within the scope of the claims of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims (8)
1. A cloth cover defect detecting and assisting device based on YOLO-v3 comprises cloth which is produced and operated on line, the cloth is conveyed by a cloth roller, the cloth roller is driven by a variable frequency motor, and the cloth cover defect detecting and assisting device is characterized in that: the device also comprises an industrial camera, an image acquisition card, a processor, a register, a PLC (programmable logic controller), a display screen, a brake block and an unwinding roller; the industrial camera is positioned above the cloth; the unwinding roller is also driven by a variable frequency motor; the brake blocks are respectively arranged on the cloth roller and the unwinding roller; the industrial camera, the image acquisition card, the processor, the register and the PLC are electrically connected, and the PLC is respectively electrically connected with the display screen and the variable frequency motor.
2. A processing method of a cloth cover defect detecting and assisting device based on YOLO-v3, according to claim 1, characterized by comprising defect inspection and cloth repair assistance, specifically:
the defect inspection comprises the following steps:
a. making a training data set and a testing data set;
b. determining the structure and training parameters of the model according to the types of the common defects, importing training data, and starting to train the weight of the model;
c. the weights of the trained models are saved and tested. The automatic monitoring of the cloth defects is realized;
d. when the defect detection device detects defects, writing the types and the position information of the defects into a system;
cloth repair assist
The cloth repair assistance comprises the following steps:
a. reading the stored defect information;
b. rewinding the cloth detected by the defect detection device to the position of the defect;
c. and the defect information is displayed on a display screen to assist in manual repair.
3. The processing method of a YOLO-v 3-based cloth cover defect detecting and assisting device as claimed in claim 2, wherein: in the defect inspection step a, defect pictures shot by an industrial camera are down-sampled and compressed to 416x416 pixels by a Gaussian pyramid, stored in common jpeg, png and bmp formats, manually screened to determine defect types and positions, and written into txt or xml format files, specifically picture names, defect types and defect coordinates.
4. The processing method of a YOLO-v 3-based cloth cover defect detecting and assisting device as claimed in claim 2, wherein: in the defect inspection step b, the input of the model is 416x416x3 three-dimensional RGB pre-processed pictures; outputting feature maps with three different sizes by a defect detection algorithm, and determining the sizes of the feature maps to be 13x13x255, 26x26x255 and 52x52x255 respectively according to the characteristics of the cloth cover defects; defects with different sizes and forms are met, fine objects can be detected by the small-scale characteristic diagram, and coarse-grained objects can be detected by the large-scale characteristic diagram; so that large, medium and small defects can be detected; after defect inspection is completed, the occurrence range of defects needs to be predicted, specifically: dividing the characteristic diagram into a grid, wherein the grid can be refined into a grid unit, each characteristic diagram corresponds to three anchor frames, characteristic diagrams with three scales are set, 9 anchor frames are provided in total, and the sizes of the 9 anchor frames are determined by a k-means clustering method according to the positions of the defects marked in advance; the prediction box has four parameters, respectively bx、by、bwAnd bh(ii) a The calculation formula is as follows (1-4):
bx＝σ(tx)+Cx (1)
by＝σ(ty)+Cy (2)
wherein t isx，tyRepresenting the predicted coordinate offset value, tw，thIs scaling; cx，CyIs the coordinates of the grid cells at the upper left corner of the feature map, and the size of each grid cell is 1x 1; p is a radical ofw，phMapping a preset anchor frame to the width and the height of the feature diagram, wherein sigma (·) represents logistic regression, and finely adjusting the prediction frame through the logistic regression;
determining the accuracy degree of the prediction box by using the IOU intersection ratio, wherein the formula is as follows (5):
where A is the prediction box and B is the true box;
determining that each grid unit has 3 kinds of anchor frames, taking the anchor frame with the largest superposition with the real frame as a prediction frame, not participating in prediction by other anchor frames, and then translating and scaling the prediction frame; the parameters of the four prediction frames are optimized by continuously training, learning and adjusting the weight, so that the intersection ratio of the prediction frames and the real frames can reach the maximum;
setting a threshold, when the intersection ratio is smaller than the threshold, the prediction is considered to be wrong and not displayed, and only when the intersection ratio is larger than the threshold, a prediction frame is displayed, wherein the threshold is set to be 0.6;
the loss function is the superposition of four parts, namely loss is made on the predicted central coordinate, loss is made on the width and height of the predicted boundary box, loss is made on the predicted category and loss is made on the predicted confidence coefficient.
5. The processing method of a YOLO-v 3-based cloth cover defect detecting and assisting device as claimed in claim 2, wherein: in the defect checking step c, the performance of the test model for detecting errors is specifically as follows: detecting the type and the predicted position of the cloth cover defects, and detecting the time length of a single picture.
6. The processing method of a YOLO-v 3-based cloth cover defect detecting and assisting device as claimed in claim 2, wherein: in the defect checking step c, the performance of the test model for detecting errors is specifically as follows: detecting the type and the predicted position of the cloth cover defects, and detecting the time length of a single picture.
7. The processing method of a YOLO-v 3-based cloth cover defect detecting and assisting device as claimed in claim 2, wherein: the reading of the stored defect information in the cloth repair assisting step a is to read the defect information and the position detected by the defect inspection device.
8. The processing method of a YOLO-v 3-based cloth cover defect detecting and assisting device as claimed in claim 2, wherein: and in the cloth repairing auxiliary step b, the detected cloth is rewound to the corresponding defect position, and the defect type is displayed on a display screen, so that the manual repairing is assisted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011111554.0A CN112215824A (en) | 2020-10-16 | 2020-10-16 | YOLO-v 3-based cloth cover defect detection and auxiliary device and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011111554.0A CN112215824A (en) | 2020-10-16 | 2020-10-16 | YOLO-v 3-based cloth cover defect detection and auxiliary device and method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112215824A true CN112215824A (en) | 2021-01-12 |
Family
ID=74055519
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011111554.0A Pending CN112215824A (en) | 2020-10-16 | 2020-10-16 | YOLO-v 3-based cloth cover defect detection and auxiliary device and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112215824A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113433137A (en) * | 2021-08-26 | 2021-09-24 | 南通三田纺织有限公司 | Textile cloth surface flaw detection device based on YOLO neural network |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103207186A (en) * | 2013-04-08 | 2013-07-17 | 北京经纬纺机新技术有限公司 | Identification method for defect detection of automatic cloth inspecting machine and system thereof |
| CN105466938A (en) * | 2015-05-05 | 2016-04-06 | 北京经纬纺机新技术有限公司 | C/S structure-based fabric defect detection information management system and method thereof |
| CN108520114A (en) * | 2018-03-21 | 2018-09-11 | 华中科技大学 | A kind of textile cloth defect detection model and its training method and application |
| CN110175658A (en) * | 2019-06-26 | 2019-08-27 | 浙江大学 | A kind of distress in concrete recognition methods based on YOLOv3 deep learning |
| CN110490874A (en) * | 2019-09-04 | 2019-11-22 | 河海大学常州校区 | Weaving cloth surface flaw detecting method based on YOLO neural network |
| CN110533640A (en) * | 2019-08-15 | 2019-12-03 | 北京交通大学 | Based on the track circuit disease discrimination method for improving YOLOv3 network model |
| CN111024720A (en) * | 2019-12-13 | 2020-04-17 | 东华大学 | Visual and tactile integrated detection device for protruding defects on surface of fabric |
| CN111210408A (en) * | 2019-12-30 | 2020-05-29 | 南京航空航天大学 | Ray image-based composite material defect identification method |
| CN111652853A (en) * | 2020-05-11 | 2020-09-11 | 西南科技大学 | Magnetic powder flaw detection method based on deep convolutional neural network |
| CN111754470A (en) * | 2020-06-11 | 2020-10-09 | 厦门雨程户外运动用品有限公司 | Automatic cloth inspecting method and device, automatic cloth inspecting machine and storage medium |
| WO2020206861A1 (en) * | 2019-04-08 | 2020-10-15 | 江西理工大学 | Yolo v3-based detection method for key object at transportation junction |
-
2020
- 2020-10-16 CN CN202011111554.0A patent/CN112215824A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103207186A (en) * | 2013-04-08 | 2013-07-17 | 北京经纬纺机新技术有限公司 | Identification method for defect detection of automatic cloth inspecting machine and system thereof |
| CN105466938A (en) * | 2015-05-05 | 2016-04-06 | 北京经纬纺机新技术有限公司 | C/S structure-based fabric defect detection information management system and method thereof |
| CN108520114A (en) * | 2018-03-21 | 2018-09-11 | 华中科技大学 | A kind of textile cloth defect detection model and its training method and application |
| WO2020206861A1 (en) * | 2019-04-08 | 2020-10-15 | 江西理工大学 | Yolo v3-based detection method for key object at transportation junction |
| CN110175658A (en) * | 2019-06-26 | 2019-08-27 | 浙江大学 | A kind of distress in concrete recognition methods based on YOLOv3 deep learning |
| CN110533640A (en) * | 2019-08-15 | 2019-12-03 | 北京交通大学 | Based on the track circuit disease discrimination method for improving YOLOv3 network model |
| CN110490874A (en) * | 2019-09-04 | 2019-11-22 | 河海大学常州校区 | Weaving cloth surface flaw detecting method based on YOLO neural network |
| CN111024720A (en) * | 2019-12-13 | 2020-04-17 | 东华大学 | Visual and tactile integrated detection device for protruding defects on surface of fabric |
| CN111210408A (en) * | 2019-12-30 | 2020-05-29 | 南京航空航天大学 | Ray image-based composite material defect identification method |
| CN111652853A (en) * | 2020-05-11 | 2020-09-11 | 西南科技大学 | Magnetic powder flaw detection method based on deep convolutional neural network |
| CN111754470A (en) * | 2020-06-11 | 2020-10-09 | 厦门雨程户外运动用品有限公司 | Automatic cloth inspecting method and device, automatic cloth inspecting machine and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| 刘娆: "卷积神经网络在纺织品缺陷检测中的应用研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅰ辑》, pages 21 - 56 * |
| 马浩鹏: "基于机器视觉的乳液泵缺陷检测方法研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅰ辑》, pages 33 - 62 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113433137A (en) * | 2021-08-26 | 2021-09-24 | 南通三田纺织有限公司 | Textile cloth surface flaw detection device based on YOLO neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111402226A (en) | Surface defect detection method based on cascade convolution neural network | |
| CN110136101B (en) | Tire X-ray defect detection method based on twinning distance comparison | |
| CN109635806B (en) | Ammeter value identification method based on residual error network | |
| CN111223093A (en) | AOI defect detection method | |
| CN111951249A (en) | Mobile phone light guide plate defect visual detection method based on multitask learning network | |
| CN108074231A (en) | A kind of magnetic sheet detection method of surface flaw based on convolutional neural networks | |
| CN106355579A (en) | Defect detecting method of cigarette carton surface wrinkles | |
| CN110490842A (en) | A kind of steel strip surface defect detection method based on deep learning | |
| CN113506239B (en) | Strip steel surface defect detection method based on cross-stage local network | |
| CN111929327A (en) | Cloth defect detection method and device | |
| CN114119591A (en) | Display screen picture quality detection method | |
| CN115775236A (en) | Surface tiny defect visual detection method and system based on multi-scale feature fusion | |
| CN115294136A (en) | Artificial intelligence-based construction and detection method for textile fabric flaw detection model | |
| CN112215824A (en) | YOLO-v 3-based cloth cover defect detection and auxiliary device and method | |
| CN116205876A (en) | Unsupervised notebook appearance defect detection method based on multi-scale standardized flow | |
| CN111754502A (en) | Method for detecting surface defects of magnetic core based on fast-RCNN algorithm of multi-scale feature fusion | |
| CN113298767A (en) | Reliable go map recognition method capable of overcoming light reflection phenomenon | |
| CN115457276B (en) | Visual detection-based winding angle detection method for high-voltage winding of power transmission and distribution transformer | |
| CN115690500A (en) | Based on improve U 2 Network instrument identification method | |
| WO2021170518A1 (en) | Automatic spreader for fabrics comprising an identification system for defects and relative method | |
| CN115239615A (en) | Cloth defect detection method based on CTPN | |
| CN114926675A (en) | Method and device for detecting shell stain defect, computer equipment and storage medium | |
| CN114092441A (en) | Product surface defect detection method and system based on dual neural network | |
| CN113205136A (en) | Real-time high-precision detection method for appearance defects of power adapter | |
| CN111968076A (en) | Real-time fabric defect detection method based on S-YOLOV3 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |