CN109454006B - Detection and classification method based on device for online detection and classification of chemical fiber spindle tripping defects - Google Patents
Detection and classification method based on device for online detection and classification of chemical fiber spindle tripping defects Download PDFInfo
- Publication number
- CN109454006B CN109454006B CN201811168402.7A CN201811168402A CN109454006B CN 109454006 B CN109454006 B CN 109454006B CN 201811168402 A CN201811168402 A CN 201811168402A CN 109454006 B CN109454006 B CN 109454006B
- Authority
- CN
- China
- Prior art keywords
- image
- area
- tripwire
- label
- connected domain
- 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.)
- Active
Links
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
- B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
- B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
- B07C5/02—Measures preceding sorting, e.g. arranging articles in a stream orientating
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
- B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
- B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
- B07C5/34—Sorting according to other particular properties
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
- B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
- B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
- B07C5/36—Sorting apparatus characterised by the means used for distribution
- B07C5/361—Processing or control devices therefor, e.g. escort memory
- B07C5/362—Separating or distributor mechanisms
-
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- 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
- G06T2207/30168—Image quality inspection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a device for detecting and grading the stumbling defect of a chemical fiber spindle on line, which comprises a spindle conveyer, an image acquisition device and a defect eliminating module which are arranged on the spindle conveyer, and a good product waiting area and a defective product waiting area which are connected with the spindle conveyer; the defect eliminating module comprises an image processing and decision-making module, a quality statistics evaluation module and a defect control module; the image acquisition device comprises closed black boxes arranged above and below the filament ingot conveying device, and an LED light source group, a CCD camera group, an image acquisition card, a digital signal processor and a memory which are arranged in each closed black box. The invention has the following beneficial effects: according to the invention, the collected filament image is processed by combining machine vision and image processing technologies, the number and morphological characteristics of tripwire and interference filament can be detected simultaneously, the detected tripwire information can be used for grading the filament with tripwire defect, and the counted interference information is used for tracing the source.
Description
Technical Field
The invention relates to the technical field of on-line detection of tripwire defects, in particular to a detection and classification method of a device based on-line detection and classification of the tripwire defects of chemical fiber yarn spindles, which can simultaneously detect tripwire and interference yarn and count the interference number and morphological characteristics of the tripwire and the interference yarn.
Background
The tripwire (also called spider web wire) is the wire which appears on the end surface of the winding bobbin, part of the wire is separated from the normal winding track, the arc is changed into the chord, and the chord length is more than 5 cm. The generation of stumble wires is closely related to the machine process adjustment, the production management and the mechanical condition, and the stumble wires can be reduced by standardizing the production management and improving the mechanical condition. The existence of the stumble wire seriously affects the appearance and unwinding performance of the winding bobbin, causes broken wires and broken ends in the post-processing process, seriously affects the production of downstream links, and is a great quality item claimed by users such as production plant reduction and the like. At present, the production lines of most chemical fiber spindle production plants adopt manual detection modes for appearance observation, but the manual detection is high in labor intensity and low in production efficiency, the manual detection can only be carried out after a fiber tube is taken off the machine, the subjective performance of people can directly influence the defect detection quality of products, and the data of manual detection cannot be accurately and timely brought into a quality management system to form production quality assessment for the whole product batch, so that the traditional manual detection mode has hysteresis, the downstream processing performance is influenced, and the real reason of the defects cannot be accurately and timely found to eliminate production and management faults. The production of chemical fiber ingots is a production process with high speed and high automation, the traditional artificial defect detection can not meet the requirement of fine production, and the online wire tripping detection system based on machine vision and image processing technology can effectively ensure the detection precision of defects, generate a product quality statistics evaluation report in real time and assist in standardizing production and management processes.
How to ensure the accurate and timely appearance on-line evaluation of the chemical fiber yarn spindle on the high-speed production line under the practical problems of various varieties, various grades and more than tens of thousands of yarn spindles produced per day of long-standing yarn reels in the chemical fiber industry becomes a problem to be solved by numerous manufacturers urgently. Since the birth of industrial robots, industrial control systems are gradually perfected, the robots can replace production workers to complete heavy and complex repetitive work, but the appearance detection cannot really break through all the time. The application number is CN201510754533.3, and the name is 'industrial raw silk filament detection device and detection method' and discloses an industrial raw silk filament on-line detection device and detection method, wherein a signal trigger mechanism is adopted to detect filaments, the device is easily interfered by waste filaments in the production of a spindle, and the detection precision cannot be guaranteed to carry out subsequent classification on the spindle. The application number CN201510883141.7 entitled "surface detection device and method" discloses an appearance detection method based on image processing, and reduces the waste silk interference by an air blowing unit, but the waste silk interference which cannot be removed by the air blowing unit still cannot be solved, and the invention only ensures detection and does not solve the problem of classification in the later period.
The vision is used as an information source for appearance detection, and how to simulate the human vision even surpass the limit of the human vision is always a problem for the computer vision to break through. Machine vision is a technology for converting an image into a digital signal to be analyzed and processed by combining an image processing technology, non-contact perception is carried out on an object through an optical device, an image of a large-range real scene is automatically acquired and interpreted, and finally information acquisition and machine control are achieved. The machine vision has the greatest advantages that the machine vision is not in contact with the detected object, so that the machine vision does not cause any damage to the observed and observed people, is safe and reliable, and does not influence the production process; moreover, machine vision in a range which can not be observed by human eyes can be observed through certain mathematical calculation, such as measurement of the area and the length of an object; in addition, the human eyes are not sensitive to the illumination conditions such as infrared rays, microwaves and ultrasonic waves, so that detailed information under the illumination conditions can be ignored, and machine vision can form specific images such as infrared rays, microwaves and ultrasonic waves by using a specific sensor to assist the human eyes to observe; in addition, human eyes cannot observe objects for a long time due to the influences of visual fatigue, subjective feeling and the like, and machine vision has no time limit and has consistent high precision and resolution; finally, the machine vision can collect and store data in production and manufacturing, and the data is analyzed and processed by a software algorithm to make a decision and a judgment, so that an intelligent and data industrial chain is required to be made, the machine is made to make a 'smart' decision, and the assistance of the machine vision is indispensable.
In recent years, with the popularization of informatization degrees of various industries, data accumulated in various industries is more and more, when massive data are faced, a traditional machine learning method is unconscious, and an efficient and accurate analysis, identification and mining method for image content is lacked, at this time, deep learning relying on large-scale data for training appears, and becomes a favorite of a machine vision algorithm field once, the appearance of the deep learning greatly shortens the research and development period while ensuring the precision, and saves the time cost for the whole production chain. Deep learning has its own advantages, but not all the more, it is sensitive to noise and influence caused by light source, focusing, shooting angle, etc., and thus there is a great instability to slightly inconsistent images, and conventional machine vision algorithms can deal well with these problems. Detection tasks can be better accomplished only by combining traditional machine vision algorithms with deep learning, such as extracting easily quantifiable features with traditional machine vision algorithms: the characteristics such as color, area, roundness, length, angle and the like are detected by deep learning, and the characteristics which are difficult to extract are as follows: flaws, defects, etc.
Disclosure of Invention
The invention provides a detection grading method of a device for detecting and grading on line based on the stumbling defect of a chemical fiber spindle, which can simultaneously detect stumbling wires and interference wires, can count the interference number and morphological characteristics of the detected stumbling wires and the interference wires, can grade the spindle with the stumbling defect by using the detected stumbling wire information, can trace the source by using the counted interference information, and standardizes the production management and the operation flow in order to overcome the defects of low accuracy, low efficiency, hysteresis of detection, serious false detection of an automatic detection system and low grading precision in the prior art.
In order to achieve the purpose, the invention adopts the following technical scheme:
a device for detecting and grading the stumbling defect of a chemical fiber spindle on line comprises a spindle conveyer, an image acquisition device and a defect eliminating module which are arranged on the spindle conveyer, and a non-defective product waiting area and a defective product waiting area which are connected with the spindle conveyer; the defect eliminating module comprises an image processing and decision-making module, a quality statistics evaluation module and a defect control module; the image acquisition device comprises closed black boxes arranged above and below the filament ingot conveying device, and a light source group, a camera group, an image acquisition card, a digital signal processor and a memory which are arranged in each closed black box; the light source group and the camera group are electrically connected with the image acquisition card, and the digital signal processor is electrically connected with the image acquisition card and the memory respectively.
The invention relates to a silk ingot conveying device, which is used for conveying silk ingots; the image acquisition device finishes the acquisition process of the label information and the image information of the silk ingot; the image processing and decision module processes and analyzes the image, finishes reading the information of the wire ingot label, detects and positions the wire tripping defect in the image of the wire ingot and obtains wire tripping detection information; the quality statistics evaluation module receives the processing results of the image processing and decision module, grades the tripwire defects according to the set grading standard to obtain quality evaluation reports corresponding to the wire ingots, and sends an eliminating instruction to the defect control module; and the defect control module receives the removing instruction to complete the removing operation of the defective silk ingots.
Preferably, the ingot conveying device comprises a main conveying belt, an auxiliary conveying belt, a conveying and shunting device, a tray and a tray fastening device; the tray is fastened on the main conveyor belt and the auxiliary conveyor belt through a tray fastening device, and the main conveyor belt and the auxiliary conveyor belt are connected through a conveying and shunting device.
A detection and classification method of a device for detecting and classifying the faults of the chemical fiber spindle stumbled yarn on line comprises the following steps:
(3-1) setting tripwire grading standards and defect levels in a quality statistics evaluation module;
(3-2) adjusting the CCD camera and the light source group to enable the defects to be visible obviously, and photographing the upper end face and the lower end face of the spindle to finish image acquisition;
(3-3) the image processing and decision-making module receives the image information transmitted by the image acquisition device, processes the image information and obtains label information and tripwire detection information;
(3-4) the quality statistics evaluation module receives the label information and tripwire detection information transmitted by the image processing and decision module, compares the obtained tripwire detection information with the set tripwire grading standard and defect grade, generates a quality evaluation report and gives a rejection instruction;
and (3-5) the defect control module receives the eliminating instruction transmitted by the quality statistics and evaluation module, transmits the defective silk ingots to a defective waiting area through the silk ingot transmission device according to the eliminating instruction, and transmits the non-defective silk ingots to a non-defective waiting area.
Preferably, the specific steps of step (3-3) are as follows:
(4-1) the image processing and decision-making module receives the image information transmitted by the image acquisition device, preprocesses the image, and positions and segments the label area in the image;
(4-2) segmenting the characters in the obtained label area, identifying the characters, and generating label information;
(4-3) segmenting the spinning cake region by combining a maximum entropy threshold segmentation method and shape fitting;
(4-4) traversing the spinning cake area through the rectangular sliding windows, and identifying local stumbled yarns and abnormal yarns for each rectangular window;
(4-5) extracting a local tripwire area of the rectangular window with the tripwire recognized by a straight line feature extraction method;
and (4-6) splicing all local tripwire identification results to obtain a global tripwire outline, and segmenting the global tripwire outline.
Preferably, the specific steps of locating and segmenting the label region in the image are as follows:
(5-1) setting a prior threshold value of the area of the connected domain as [ Ami, Ama ];
(5-2) obtaining an adaptive threshold value t by an Otsu Daohui method, and calculating a binary image of the image collected by the image collection device under the threshold value t;
(5-3) calculating all independent connected domains of the obtained binary image, judging the area of each connected domain, taking the connected domain with the area size within the range of prior threshold value [ Ami, Ama ] as an optimal connected domain, and cutting the minimum contained rectangle of the optimal connected domain to obtain a label region;
(5-4) correcting the rectangular inclination of the label area, calculating the inclination angle of the label area by using Hough transform on the binary image, and then correcting the rectangular inclination of the label area by image rotation, wherein the rotation formula is as follows:
wherein, (x, y) is the coordinate of the middle point of the original image, (x ', y') is the coordinate of the corresponding point in the rotated image, and theta is the rotation angle;
(5-5) correcting perspective distortion of the label area, searching an image outsourcing rectangle for the binary image, calculating outline corner points of the image outsourcing rectangle, selecting four corner points of the image outsourcing rectangle, calculating to obtain a projection transformation matrix through the four corner points of the image outsourcing rectangle and four vertexes of the image outsourcing rectangle, and obtaining an image after perspective transformation correction through the projection transformation matrix; wherein the transformation formula is:
wherein, (u, v) is the coordinate of the central point in the original image, and (u ', v') is the coordinate of the corresponding point in the image after projection transformation, and the matrix
Is a projective transformation matrix.
Preferably, the specific steps of step (4-2) are as follows:
(6-1) setting the prior gray threshold range to be sa，sb]The Area threshold range is [ Arei, Area]The distance threshold range is [ dsi, dsa]The angle range is [ ani, ana]；
(6-2) MSER character candidate region extraction using [0, 1, 2 … 255]The gray threshold value of the image is subjected to binarization processing, corresponding black and white areas are obtained for the binary image obtained by each threshold value, and a priori gray threshold value range [ s ] is extracteda，sb]The region with the stable shape is the most stable extremum region, the most stable extremum region is extracted from the MSER character candidate region, wherein the region with the stable shape is as follows:
where A denotes the area of the region of the binary image, th denotes the grayscale threshold, dARepresenting the amount of change in area of a region of a binary image, dthRepresenting the amount of change in the gray threshold.
(6-3) character segmentation;
(6-3-1) calculating the Area Ar of the independent connected domain for the MSER character candidate region, and if the Area Ar of the connected domain meets the condition that Ar is not less than Arei and not more than Area, turning to the step (6-3-2);
if the Area Ar of the connected domain meets Ar < Arei or Ar > Area, switching to the step (6-3-5);
(6-3-2) calculating the distance ds from the connected domain to the [0, 0] pixel point for the MSER character candidate region, and if the distance ds from the connected domain to the [0, 0] pixel point meets the condition that dsi is not less than ds and not more than dsa, turning to the step (6-3-3);
if the distance ds from the connected domain to the [0, 0] pixel point meets ds < dsi or ds > dsa, the step (6-3-5) is carried out;
(6-3-3) calculating the angle an of the minimum circumscribed rectangle of the connected domain for the MSER character candidate region, and if the angle an of the minimum circumscribed rectangle of the connected domain meets the condition that ani is not less than an and not more than ana, turning to the step (6-3-4);
if the angle an of the minimum circumscribed rectangle of the connected domain meets an & ltani or an & gt ana, turning to the step (6-3-5);
(6-3-4) solving minimum contained rectangular frames of the connected domain, wherein each minimum contained rectangular frame contains a character, and intercepting the minimum contained rectangular frames from the acquired image information to be used as candidate data sets for character recognition;
(6-3-5) discarding the connected domain;
(6-4) training a character recognizer, collecting label data of all collected images, extracting all minimum inclusion rectangular frames, carrying out size normalization processing on all the minimum inclusion rectangular frames, endowing each minimum inclusion rectangular frame with a corresponding character label, sending the label data and the character labels into a cifar-10 network with a pre-training model, and carrying out character recognizer training to obtain character segmentation parameters;
(6-5) splicing the character segmentation results, processing the acquired test image through a character recognizer, obtaining the character segmentation results according to the character segmentation parameters, and splicing the character segmentation results according to the coordinate information of the character segmentation results to generate label information.
Preferably, the specific steps of step (4-3) are as follows:
(7-1) carrying out image smoothing treatment, carrying out gray processing on the image subjected to image preprocessing, and then carrying out neighborhood averaging on the gray image to obtain a smooth image;
(7-2) calculating a two-dimensional histogram, wherein a two-dimensional histogram is constructed by the image information and the smooth image transmitted by the image acquisition device;
(7-3) carrying out image binarization, searching an optimal threshold value according to a maximum entropy principle, and carrying out binarization processing on the smooth image by using the threshold value;
and (7-4) arc fitting, calculating all independent connected domains in the binarized image, calculating the connected domain with the largest area as the optimal connected domain, then carrying out arc fitting on the optimal connected domain outline point set, and extracting the arc region from the image, namely the spinning cake region.
Preferably, the specific steps of step (4-4) are as follows:
(8-1) setting the size of the cake area obtained in the step (7-4) to Ls×DsThe size of the rectangular window is Lj×Dj，
(8-2) size Lj×DjPerforming sliding window traversal on the spinning cake area obtained in the step (7-4) by using the rectangular window to obtain n multiplied by m rectangular windows;
(8-3) performing sliding window traversal on all the acquired image information to obtain all rectangular windows, labeling each rectangular window, wherein each rectangular window corresponds to two labels, one is whether an abnormal thread has a label or not, and the other is whether a thread is tripped or not, and sending the label labeling information of all the rectangular windows into a VGG network with a pre-training weight to perform multi-label classifier training to obtain label classification parameters;
(8-4) performing rectangular window sliding traversal on the collected filament ingot image for testing to obtain all rectangular windows; and sending each rectangular window into a multi-label classifier for classification, recording abnormal silk thread information if the label is abnormal silk and no stumbled silk, and outputting the rectangular window if the label is abnormal silk and stumbled silk.
Preferably, the specific steps of step (4-5) are as follows:
(9-1) setting the length range of the rectangular frame as [ L1, L2], the angle range as [ ani1, ana1], and the Area range as [ Arei1, Area1 ];
(9-2) performing Gaussian kernel sampling on the rectangular window with the stumbled wire identified in the step (8-4);
(9-3) calculating the gradient value and the gradient direction of each point in the rectangular window, and then sequencing all the points according to the gradient values;
(9-4) judging each point in the sorted point set, eliminating pixel points with gradient amplitudes smaller than a set threshold value, and combining pixel points with similar directions in the rest pixel points by a region growing method;
(9-5) performing rectangle estimation on the discrete point set obtained by the region growing method, including all discrete points in a rectangle box, and calculating the length L, the angle an1 and the area A1 of the rectangle box; rectangular boxes satisfying L1-L2, ani 1-an 1-ana 1 and Arei 1-A1-Area 1 are extracted as line segments to serve as local wire-tripping regions.
Preferably, the specific steps of step (4-6) are as follows:
(10-1) setting a length prior threshold L2 of a connected domain
Is Ratio;
(10-2) splicing the rectangular frames obtained in the step (9-5) according to the coordinate information of the rectangular frames to obtain a wire tripping mask image corresponding to a wire cake area;
(10-3) carrying out morphological corrosion operation on the tripwire mask image to obtain a corroded mask image;
(10-4) carrying out independent connected domain calculation on the corroded mask image, calculating the length and the area of each connected domain, filling the connected domains with the lengths smaller than a prior threshold L2 into black,
is greater thanFilling black in a connected domain of the prior threshold value Ratio to obtain a new mask image;
(10-5) counting the length of each connected domain, the number of the connected domains and the relative position information of the connected domains in the new mask image generated in the step (10-4), taking the length of the connected domains as the tripwire length, taking the number of the connected domains as the tripwire number and taking the relative position information of the connected domains as the position information of the tripwire.
Therefore, the invention has the following beneficial effects: (1) the whole detection device does not influence the normal production of the chemical fiber spindle, and realizes the purposes of collecting images of the chemical fiber spindle, removing defective products and feeding back a whole set of coherent online processing flow to detection; (2) the invention utilizes the image processing technology to extract the label information, avoids the installation of redundant hardware equipment, saves the cost, defines the 'identity' of the silk ingot by taking a picture once and completes the defect detection of the silk ingot; (3) according to the invention, by utilizing the motion rule of the conveyor belt and installing the conveying and shunting device, defective silk ingots can be removed timely, accurately and at high speed; (4) the invention has a special quality statistic evaluation module, and the quality statistic evaluation module can accurately grade the defects according to the defect information only by importing the defect evaluation standard, thereby easily solving the problem of non-uniform standards caused by different production plants and different silk ingot varieties; (5) according to the invention, the collected filament ingot images are processed by combining machine vision and an image processing technology, so that the tripwire and the interference wire can be detected simultaneously, meanwhile, the interference number and the morphological characteristics of the tripwire and the interference wire can be counted and detected, the filament ingots with tripwire defects can be graded by using the detected tripwire information, the counted interference information is used for tracing the origin, and the production management and the operation flow are standardized.
Drawings
FIG. 1 is a schematic diagram of an embodiment of the present invention;
FIG. 2 is a block diagram of an image capture device of the present invention;
FIG. 3 is a flow chart of the present invention;
FIG. 4 is a flow chart of a method for locating and segmenting tag information according to the present invention;
FIG. 5 is a schematic flow diagram of a tripwire defect detection and location method of the present invention;
FIG. 6 is a schematic flow diagram of a method of segmenting a cake region of a spinning ingot in accordance with the present invention;
FIG. 7 is a schematic flow chart of the linear feature extraction method of the present invention;
FIG. 8 is a schematic flow diagram of the quality statistic evaluation module of the present invention;
FIG. 9 is a flow chart of a defect control module according to the present invention.
In the figure: the device comprises a silk ingot conveying device 1, an image acquisition device 2, a defect eliminating module 3, a non-defective product waiting area 4, a defective product waiting area 5, a main conveying belt 11, an auxiliary conveying belt 12, a conveying and shunting device 13, a tray 14, a light source group 21, a camera group 22, an image acquisition card 23, a digital signal processor 24, a memory 25, an image processing and decision-making module 31, a quality statistics evaluation module 32 and a defect control module 33.
Detailed Description
The invention is further described in the following detailed description with reference to the drawings in which:
the embodiment shown in fig. 1 is an on-line detection and classification device for the tripping defect of a chemical fiber spindle, which comprises a spindle conveyer 1, an image acquisition device 2 and a defect removal module 3 which are arranged on the spindle conveyer, a good product waiting area 4 and a defective product waiting area 5 which are connected with the spindle conveyer; the defect eliminating module comprises an image processing and decision-making module 31, a quality statistics evaluation module 32 and a defect control module 33; the wire ingot conveying device comprises a main conveying belt 11, an auxiliary conveying belt 12, a conveying and shunting device 13, a tray 14 and a tray fastening device; the tray is fastened on the main conveyor belt and the auxiliary conveyor belt through a tray fastening device, and the main conveyor belt and the auxiliary conveyor belt are connected through a conveying and shunting device.
The tray is provided with a tray for loading the ingots, the tray fastening device is provided with a loading and unloading valve, the tray can be fastened on the main conveyor belt and the auxiliary conveyor belt through a fastening switch, the loading and unloading of the tray are realized through the loading and unloading valve, the tray is used for bearing the ingots, a support rod is arranged in the center of the tray, an elastic gripper is arranged on the support rod and used for stabilizing the ingots in the center of the tray, and the phenomenon that the tray shakes to influence the imaging and conveying of the ingots in the conveying process of the main conveyor belt and the auxiliary conveyor belt is avoided; the main conveyor belt conveys the filament ingots to pass through the image acquisition device to complete image acquisition, when the defect control module does not receive an eliminating instruction, the filament ingots reach a non-defective product waiting area, when the defect control module receives the eliminating instruction, the conveying and shunting device is started to shunt the defective filament ingots to the auxiliary conveyor belt, and the auxiliary conveyor belt conveys the defective filament ingots to a defective product waiting area.
As shown in fig. 2, the image capturing device includes closed black boxes disposed above and below the filament conveying device, and a light source group 21, a camera group 22, an image capturing card 23, a digital signal processor 24 and a memory 25 disposed in each closed black box; the light source group and the camera group are electrically connected with the image acquisition card, and the digital signal processor is electrically connected with the image acquisition card and the memory respectively.
A photoelectric sensor capable of acquiring high-precision images is arranged in the CCD camera, and images of moving objects in a stable time period are acquired by combining an external trigger scanning mode and controllable exposure time; the light source groups are arranged at the upper and lower proper positions of the image acquisition device, so that the influence of vibration on the angle of the light source is prevented, the filament ingot is stably polished, and CCD cameras are arranged at proper positions and directions of the upper end and the bottom end of the image acquisition device; when the filament ingot reaches the exposure range of the CCD camera, the light reflected by the filament ingot is projected onto the sensor through the camera lens, after the sensor is exposed, the photodiode is excited by light to release charges, electric signals are generated, the electric signals control the current generated by the photodiode through a camera chip by utilizing a control signal circuit in a photosensitive element and are output by a current transmission circuit, the camera chip collects the electric signals generated by primary imaging and uniformly outputs the electric signals to an amplifier, the electric signals after amplification and filtering are sent to an A/D (analog/digital) converter, the A/D converter converts the electric signals into digital signals, and outputting the images to a digital signal processor, carrying out post-image processing such as color correction and white balance processing on the images by the digital signal processor, encoding the images into image files with specific resolution and image format supported by DC, and finally saving the image files to a memory.
A detection and classification method of a device for detecting and classifying the faults of the chemical fiber spindle stumbled yarn on line is shown in figure 3, and comprises the following steps:
step 100, setting tripwire grading standards and defect grades in a quality statistics evaluation module;
step 200, adjusting the CCD camera and the light source group to enable the defects to be visible obviously, and photographing the upper end face and the lower end face of the spindle to finish image acquisition;
step 300, an image processing and decision-making module receives image information transmitted by an image acquisition device, processes the image information and obtains label information and tripwire detection information;
step 301, an image processing and decision-making module receives image information transmitted by an image acquisition device, preprocesses an image, and positions and segments a label area in the image;
step 302, segmenting the characters in the obtained label area, identifying the characters and generating label information;
303, segmenting the spinning cake region by combining a maximum entropy threshold segmentation method and shape fitting;
304, traversing the spinning cake area through the rectangular sliding windows, and identifying local stumbles and abnormal silk threads in each rectangular window;
305, extracting a local tripwire area of a rectangular window with the tripwire recognized by a straight line feature extraction method;
step 306, splicing all local tripwire recognition results to obtain a global tripwire outline, and segmenting the global tripwire outline;
400, a quality statistics evaluation module receives label information and tripwire detection information transmitted by an image processing and decision module, compares the obtained tripwire detection information with a set tripwire grading standard and a set defect grade, generates a quality evaluation report and gives a rejection instruction;
and 500, receiving a rejection instruction transmitted by the quality statistics and evaluation module by the defect control module, transmitting the defective silk ingots to a defective waiting area through a silk ingot transmission device according to the rejection instruction, and transmitting the non-defective silk ingots to a non-defective waiting area.
The label of the spindle is the 'ID card' of the spindle, the spindle label is accurately read, and the tracing of the chemical fiber spindle from the coming to the coming can be realized; acquiring tag information by processing an image acquired from a camera directly above an image acquisition device, the tag information identification including the steps of: positioning a label area, segmenting characters, identifying the characters and outputting a label result; since the background of the label is completely white, the text is completely black, and the label area and other areas of the image have large color difference, the label area is segmented by using a thresholding method, as shown in fig. 4, the specific steps from step 301 to step 302 are as follows:
(5-1) setting a priori threshold value of the area of the connected domain as [ Ami, Ama]The prior gray threshold range is [ s ]a，sb]The Area threshold range is [ Arei, Area]The distance threshold range is [ dsi, dsa]The angle range is [ ani, ana]；
(5-2) obtaining an adaptive threshold value t by an Otsu Daohui method, and calculating a binary image of the image collected by the image collection device under the threshold value t;
(5-3) calculating all independent connected domains of the obtained binary image, judging the area of each connected domain, taking the connected domain with the area size within the range of prior threshold value [ Ami, Ama ] as an optimal connected domain, and cutting the minimum contained rectangle of the optimal connected domain to obtain a label region;
(5-4) correcting the rectangular inclination of the label area, calculating the inclination angle of the label area by using Hough transform on the binary image, and then correcting the rectangular inclination of the label area by image rotation, wherein the rotation formula is as follows:
wherein, (x, y) is the coordinate of the middle point of the original image, (x ', y') is the coordinate of the corresponding point in the rotated image, and theta is the rotation angle;
(5-5) correcting perspective distortion of the label area, searching an image outsourcing rectangle for the binary image, calculating outline corner points of the image outsourcing rectangle, selecting four corner points of the image outsourcing rectangle, calculating to obtain a projection transformation matrix through the four corner points of the image outsourcing rectangle and four vertexes of the image outsourcing rectangle, and obtaining an image after perspective transformation correction through the projection transformation matrix; wherein the transformation formula is:
wherein, (u, v) is the coordinate of the central point in the original image, and (u ', v') is the coordinate of the corresponding point in the image after projection transformation, and the matrix
A projective transformation matrix;
(5-6) MSER character candidate region extraction using [0, 1, 2 … 255]The gray threshold value of the image is subjected to binarization processing, corresponding black and white areas are obtained for the binary image obtained by each threshold value, and a priori gray threshold value range [ s ] is extracteda，sb]The region with the stable shape is the most stable extremum region, the most stable extremum region is extracted from the MSER character candidate region, wherein the region with the stable shape is as follows:
where A denotes the area of the region of the binary image, th denotes the grayscale threshold, dARepresenting the amount of change in area of a region of a binary image, dthRepresenting the amount of change in the gray threshold.
(5-7) character segmentation;
(5-7-1) calculating the Area Ar of the independent connected domain for the MSER character candidate region, and if the Area Ar of the connected domain meets the condition that Ar is not less than Arei and not more than Area, turning to the step (5-7-2);
if the Area Ar of the connected domain meets Ar < Arei or Ar > Area, turning to the step (5-7-5);
(5-7-2) calculating the distance ds from the connected domain to the [0, 0] pixel point for the MSER character candidate region, and if the distance ds from the connected domain to the [0, 0] pixel point meets the condition that dsi is not less than ds and not more than dsa, turning to the step (5-7-3);
if the distance ds from the connected domain to the [0, 0] pixel point meets ds < dsi or ds > dsa, the step (5-7-5) is carried out;
(5-7-3) calculating the angle an of the minimum circumscribed rectangle of the connected domain for the MSER character candidate region, and if the angle an of the minimum circumscribed rectangle of the connected domain meets the condition that ani is not less than an and not more than ana, turning to the step (5-7-4);
if the angle an of the minimum circumscribed rectangle of the connected domain meets an & lt ani or an & gt ana, turning to the step (5-7-5);
(5-7-4) solving minimum contained rectangular frames of the connected domain, wherein each minimum contained rectangular frame contains a character, and intercepting the minimum contained rectangular frames from the acquired image information to be used as candidate data sets for character recognition;
(5-7-5) discarding the connected domain;
(5-8) training a character recognizer, collecting label data of all collected images, extracting all minimum inclusion rectangular frames, carrying out size normalization processing on all the minimum inclusion rectangular frames, endowing each minimum inclusion rectangular frame with a corresponding character label, sending the label data and the character labels into a cifar-10 network with a pre-training model, and carrying out character recognizer training to obtain character segmentation parameters;
(5-9) splicing the character segmentation results, processing the acquired test image through a character recognizer, obtaining the character segmentation results according to the character segmentation parameters, and splicing the character segmentation results according to the coordinate information of the character segmentation results to generate label information.
The tripping wire is mostly generated due to improper operation of workers or process reasons such as faults of production apparatuses, so that the tripping wire on a chemical fiber spindle is indefinite in length, extending direction and shape and has no certain regularity, and the phenomenon is relatively troublesome in the field of computer vision; due to the particularity of the tripwire, the invention adopts the thought from local to global to solve the problem; as shown in fig. 5, the tripwire defect identification and location is primarily divided into the following steps: image preprocessing, spinning cake area segmentation, wire tripping identification and global wire tripping contour positioning; industrial images are often affected by optical, electrical, electromechanical movement, component materials, and circuits of devices inside the system during generation to generate noise, and the images are also often accompanied by noise generation during digitization and transmission, and the presence of noise not only affects the visual effect of the images, but also affects many image processing operations such as: edge detection, image segmentation and the like can generate more or less influence, so that the image preprocessing stage mainly focuses on the suppression of image noise, and the median filter is used for denoising the image; the position where the tripwire appears is a spinning cake area of a spinning spindle, and the existence of other areas except the spinning cake can increase the image processing time and interfere the image processing result, so that the spinning cake area needs to be divided from the image, and the concentration degree of image processing is increased; as shown in fig. 6, the specific steps of step 303 are as follows:
(7-1) carrying out image smoothing treatment, carrying out gray processing on the image subjected to image preprocessing, and then carrying out neighborhood averaging on the gray image to obtain a smooth image;
(7-2) calculating a two-dimensional histogram, wherein a two-dimensional histogram is constructed by the image information and the smooth image transmitted by the image acquisition device;
(7-3) carrying out image binarization, searching an optimal threshold value according to a maximum entropy principle, and carrying out binarization processing on the smooth image by using the threshold value;
and (7-4) arc fitting, calculating all independent connected domains in the binarized image, calculating the connected domain with the largest area as the optimal connected domain, then carrying out arc fitting on the optimal connected domain outline point set, and extracting the arc region from the image, namely the spinning cake region.
The specific steps of step 304 are as follows:
(8-1) setting the size of a spinning cake area to be Ls×DsThe size of the rectangular window is Lj×Dj，
(8-2) size Lj×DjPerforming sliding window traversal on the spinning cake area obtained in the step (7-4) by using the rectangular window to obtain n multiplied by m rectangular windows;
(8-3) performing sliding window traversal on all the acquired image information to obtain all rectangular windows, labeling each rectangular window, wherein each rectangular window corresponds to two labels, one is whether an abnormal thread has a label or not, and the other is whether a thread is tripped or not, and sending the label labeling information of all the rectangular windows into a VGG network with a pre-training weight to perform multi-label classifier training to obtain label classification parameters;
(8-4) performing rectangular window sliding traversal on the collected filament ingot image for testing to obtain all rectangular windows; and sending each rectangular window into a multi-label classifier for classification, recording abnormal silk thread information if the label is abnormal silk and no stumbled silk, and outputting the rectangular window if the label is abnormal silk and stumbled silk.
As shown in fig. 7, the specific steps of step 305 are as follows:
(9-1) setting the length range of the rectangular frame as [ L1, L2], the angle range as [ ani1, ana1], and the Area range as [ Arei1, Area1 ];
(9-2) performing Gaussian kernel sampling on the rectangular window with the stumbled wire identified in the step (8-4);
(9-3) calculating the gradient value and the gradient direction of each point in the rectangular window, and then sequencing all the points according to the gradient values;
(9-4) judging each point in the sorted point set, eliminating pixel points with gradient amplitudes smaller than a set threshold value, and combining pixel points with similar directions in the rest pixel points by a region growing method;
(9-5) performing rectangle estimation on the discrete point set obtained by the region growing method, including all discrete points in a rectangle box, and calculating the length L, the angle an1 and the area A1 of the rectangle box; rectangular boxes satisfying L1-L2, ani 1-an 1-ana 1 and Arei 1-A1-Area 1 are extracted as line segments to serve as local wire-tripping regions.
The specific steps of step 306 are as follows:
(10-1) setting a length prior threshold L2 of a connected domain
Is Ratio;
(10-2) splicing the rectangular frames obtained in the step (9-5) according to the coordinate information of the rectangular frames to obtain a wire tripping mask image corresponding to a wire cake area;
(10-3) carrying out morphological corrosion operation on the tripwire mask image to obtain a corroded mask image;
(10-4) carrying out independent connected domain calculation on the corroded mask image, calculating the length and the area of each connected domain, filling the connected domains with the lengths smaller than a prior threshold L2 into black,
filling black in the connected domain larger than the prior threshold value Ratio to obtain a new mask image;
(10-5) counting the length of each connected domain, the number of the connected domains and the relative position information of the connected domains in the new mask image generated in the step (10-4), taking the length of the connected domains as the tripwire length, taking the number of the connected domains as the tripwire number and taking the relative position information of the connected domains as the position information of the tripwire.
Fig. 8 is a functional diagram of a quality statistics evaluation module, which is mainly used for performing defect information statistics and management to obtain a final evaluation report, and the main operation details are the establishment of a database; the defect information statistics mainly comprises daily produced silk ingot total quantity statistics, daily inspection tripwire quantity statistics, daily defect rejection quantity statistics and batch abnormal silk statistics, wherein the daily produced silk ingot total quantity statistics can carry out real-time monitoring on a production chain, the daily tripwire quantity statistics can assist in evaluating production and management processes, and the daily defect rejection quantity statistics, the batch abnormal silk statistics and the like can evaluate the production quality; after each filament passes through the image acquisition device, the quality statistics and evaluation module receives label information, defect information, abnormal filament information and time information of the filament arriving at the defect removal module, can count the total quantity of filament produced in daily life according to the received label information, compares stumble filament classification standards in the evaluation module according to the received defect information, can classify stumble filament defects, counts the number of stumble filaments detected in daily life, obtains a quality evaluation report according to the counted number of stumble filaments and the corresponding defect grade, then generates a removal instruction corresponding to the filament, and sends the removal instruction and the removal time point to the defect control module.
The calculation method of the time for the filament ingot to reach the defect eliminating module comprises the following steps: assuming that the movement of the filament ingots on the conveyor belt is uniform, the time for the filament ingots to reach the defect removing module is as follows:
wherein s is the distance from the image acquisition device to the defect eliminating module, and v is the conveying speed of the silk ingots in the distance.
As shown in fig. 9, the defect control module is located in the defect eliminating module, and the main function of the defect control module is to receive the eliminating instruction and the eliminating time point of the quality evaluation module, and add the eliminating instruction and the eliminating time point into the instruction storage queue, if an ingot filament reaches the defect eliminating module at a certain eliminating time point, the corresponding eliminating instruction is "not to be eliminated", the corresponding eliminating instruction and the eliminating time point are deleted from the eliminating queue, the non-defective filament ingot is transmitted to the good product waiting area through the main conveyor belt, if the corresponding eliminating instruction is "to be eliminated", the defect control module starts the transmission and distribution device, the filament ingot to be eliminated is distributed to the auxiliary conveyor belt from the main conveyor belt, and the defective filament ingot is transmitted to the defective product waiting area through the auxiliary conveyor belt.
It should be understood that this example is for illustrative purposes only and is not intended to limit the scope of the present invention. Further, it should be understood that various changes or modifications of the present invention may be made by those skilled in the art after reading the teaching of the present invention, and such equivalents may fall within the scope of the present invention as defined in the appended claims.
Claims (8)
1. A detection and classification method based on a device for detecting and classifying the chemical fiber spindle tripping defects on line is characterized in that the device for detecting and classifying the chemical fiber spindle tripping defects on line comprises a spindle conveying device (1), an image acquisition device (2) and a defect eliminating module (3) which are arranged on the spindle conveying device, a good product waiting area (4) and a defective product waiting area (5) which are connected with the spindle conveying device; the defect eliminating module comprises an image processing and decision-making module (31), a quality statistics evaluation module (32) and a defect control module (33); the image acquisition device comprises closed black boxes arranged above and below the filament ingot conveying device, and a light source group (21), a camera group (22), an image acquisition card (23), a digital signal processor (24) and a memory (25) which are arranged in each closed black box; the light source group and the camera group are electrically connected with the image acquisition card, and the digital signal processor is electrically connected with the image acquisition card and the memory respectively; the method comprises the following steps:
(1-1) setting tripwire grading standards and defect levels in a quality statistics evaluation module;
(1-2) adjusting the camera and the light source group to enable the defects to be visible obviously, and photographing the upper end face and the lower end face of the spindle to finish image acquisition;
(1-3) the image processing and decision-making module receives the image information transmitted by the image acquisition device, processes the image information and obtains label information and tripwire detection information;
(1-3-1) the image processing and decision-making module receives image information transmitted by the image acquisition device, preprocesses the image, and positions and segments a label area in the image;
(1-3-2) segmenting the characters in the obtained label area, identifying the characters, and generating label information;
(1-3-3) segmenting the spinning cake region by combining a maximum entropy threshold segmentation method and shape fitting;
(1-3-4) traversing a spinning cake area through rectangular sliding windows, and identifying local stumbled yarns and abnormal yarns for each rectangular window;
(1-3-5) extracting a local tripwire area of a rectangular window with the tripwire recognized by a straight line feature extraction method;
(1-3-6) splicing all local tripwire identification results to obtain a global tripwire outline, and segmenting the global tripwire outline;
(1-4) the quality statistics evaluation module receives the label information and tripwire detection information transmitted by the image processing and decision module, compares the obtained tripwire detection information with the set tripwire grading standard and defect grade, generates a quality evaluation report and gives a rejection instruction;
(1-5) the defect control module receives the eliminating instruction transmitted by the quality statistics and evaluation module, transmits the defective silk ingots to a defective waiting area through the silk ingot transmission device according to the eliminating instruction, and transmits the non-defective silk ingots to a non-defective waiting area.
2. The detection and classification method based on the chemical fiber yarn spindle tripping defect online detection and classification device is characterized in that the yarn spindle conveying device comprises a main conveying belt (11), an auxiliary conveying belt (12), a conveying and shunting device (13), a tray (14) and a tray fastening device; the tray is fastened on the main conveyor belt and the auxiliary conveyor belt through a tray fastening device, and the main conveyor belt and the auxiliary conveyor belt are connected through a conveying and shunting device.
3. The method for detecting and grading the device based on the chemical fiber spindle tripwire defect online detection and grading according to claim 1, wherein the specific steps of positioning and dividing the label area in the image are as follows:
(3-1) setting a prior threshold value of the area of the connected domain as [ Ami, Ama ];
(3-2) obtaining an adaptive threshold value t by an Otsu Daohui method, and calculating a binary image of the image collected by the image collection device under the threshold value t;
(3-3) calculating all independent connected domains of the obtained binary image, judging the area of each connected domain, taking the connected domain with the area size within the range of prior threshold value [ Ami, Ama ] as an optimal connected domain, and cutting the minimum contained rectangle of the optimal connected domain to obtain a label region;
(3-4) correcting the rectangular inclination of the label area, calculating the inclination angle of the label area by using Hough transform on the binary image, and then correcting the rectangular inclination of the label area by image rotation, wherein the rotation formula is as follows:
wherein, (x, y) is the coordinate of the middle point of the original image, (x ', y') is the coordinate of the corresponding point in the rotated image, and theta is the rotation angle;
(3-5) correcting perspective distortion of the label area, searching an image outsourcing rectangle for the binary image, calculating outline corner points of the image outsourcing rectangle, selecting four corner points of the image outsourcing rectangle, calculating to obtain a projection transformation matrix through the four corner points of the image outsourcing rectangle and four vertexes of the image outsourcing rectangle, and obtaining an image after perspective transformation correction through the projection transformation matrix; wherein the transformation formula is:
wherein, (u, v) is the coordinate of the central point in the original image, and (u ', v') is the coordinate of the corresponding point in the image after projection transformation, and the matrix
Is a projective transformation matrix.
4. The detection and classification method based on the chemical fiber yarn spindle and yarn tripwire defect online detection and classification device according to claim 1, characterized in that the specific steps of the steps (1-3-2) are as follows:
(4-1) setting the prior gray threshold range to be sa，sb]The Area threshold range is [ Arei, Area]The distance threshold range is [ dsi, dsa]The angle range is [ ani, ana]；
(4-2) MSER character candidate region extraction using [0, 1, 2 … 255]The gray threshold value of the image is subjected to binarization processing, corresponding black and white areas are obtained for the binary image obtained by each threshold value, and a priori gray threshold value range [ s ] is extracteda，sb]The region with the stable shape is the most stable extremum region, the most stable extremum region is extracted from the MSER character candidate region, wherein the region with the stable shape is as follows:
where A denotes the area of the region of the binary image, th denotes the grayscale threshold, dARepresenting the amount of change in area of a region of a binary image, dthRepresenting a gray threshold variation;
(4-3) character segmentation;
(4-3-1) calculating the Area Ar of the independent connected domain for the MSER character candidate region, and if the Area Ar of the connected domain meets the condition that Ar is not less than Arei and not more than Area, turning to the step (4-3-2);
if the Area Ar of the connected domain meets Ar < Arei or Ar > Area, switching to the step (4-3-5);
(4-3-2) calculating the distance ds from the connected domain to the [0, 0] pixel point for the MSER character candidate region, and if the distance ds from the connected domain to the [0, 0] pixel point meets the condition that dsi is not less than ds and not more than dsa, turning to the step (4-3-3);
if the distance ds from the connected domain to the [0, 0] pixel point meets ds < dsi or ds > dsa, the step (4-3-5) is carried out;
(4-3-3) calculating the angle an of the minimum circumscribed rectangle of the connected domain for the MSER character candidate region, and if the angle an of the minimum circumscribed rectangle of the connected domain meets the condition that ani is not less than an and not more than ana, turning to the step (4-3-4);
if the angle an of the minimum circumscribed rectangle of the connected domain meets an & ltani or an & gt ana, turning to the step (4-3-5);
(4-3-4) solving minimum contained rectangular frames of the connected domain, wherein each minimum contained rectangular frame contains a character, and intercepting the minimum contained rectangular frames from the acquired image information to be used as candidate data sets for character recognition;
(4-3-5) discarding the connected domain;
(4-4) training a character recognizer, collecting label data of all collected images, extracting all minimum inclusion rectangular frames, carrying out size normalization processing on all the minimum inclusion rectangular frames, endowing each minimum inclusion rectangular frame with a corresponding character label, sending the label data and the character labels into a cifar-10 network with a pre-training model, and carrying out character recognizer training to obtain character segmentation parameters;
(4-5) splicing the character segmentation results, processing the acquired test image through a character recognizer, obtaining the character segmentation results according to the character segmentation parameters, and splicing the character segmentation results according to the coordinate information of the character segmentation results to generate label information.
5. The detection and classification method based on the chemical fiber yarn spindle and yarn tripwire defect online detection and classification device according to claim 1, characterized in that the specific steps of the steps (1-3-3) are as follows:
(5-1) carrying out image smoothing treatment, carrying out gray level treatment on the image subjected to image preprocessing, and then carrying out neighborhood averaging on the gray level image to obtain a smooth image;
(5-2) calculating a two-dimensional histogram, wherein a two-dimensional histogram is constructed by the image information and the smooth image transmitted by the image acquisition device;
(5-3) carrying out image binarization, searching an optimal threshold value according to a maximum entropy principle, and carrying out binarization processing on the smooth image by using the threshold value;
and (5-4) arc fitting, namely calculating all independent connected domains in the binarized image, calculating the connected domain with the largest area as the optimal connected domain, then carrying out arc fitting on the optimal connected domain outline point set, and extracting the arc region from the image, namely the spinning cake region.
6. The detection and classification method based on the chemical fiber yarn spindle and yarn tripwire defect online detection and classification device according to claim 5, characterized in that the specific steps of the steps (1-3-4) are as follows:
(6-1) setting the size of the cake area obtained in the step (5-4) to Ls×DsThe size of the rectangular window is Lj×Dj，
(6-2) size Lj×DjPerforming sliding window traversal on the spinning cake area obtained in the step (5-4) by using the rectangular window to obtain n multiplied by m rectangular windows;
(6-3) performing sliding window traversal on all the acquired image information to obtain all rectangular windows, labeling each rectangular window, wherein each rectangular window corresponds to two labels, one is whether an abnormal thread has a label or not, and the other is whether a thread is tripped or not, and sending the label labeling information of all the rectangular windows into a VGG network with a pre-training weight to perform multi-label classifier training to obtain label classification parameters;
(6-4) performing rectangular window sliding traversal on the collected filament ingot image for testing to obtain all rectangular windows; and sending each rectangular window into a multi-label classifier for classification, recording abnormal silk thread information if the label is abnormal silk and no stumbled silk, and outputting the rectangular window if the label is abnormal silk and stumbled silk.
7. The detection and classification method based on the chemical fiber yarn spindle and yarn tripwire defect online detection and classification device according to claim 1, characterized in that the specific steps of the steps (1-3-5) are as follows:
(7-1) setting the length range of the rectangular frame as [ L1, L2], the angle range as [ ani1, ana1], and the Area range as [ Arei1, Area1 ];
(7-2) performing Gaussian kernel sampling on the rectangular window with the stumbled wire identified in the step (6-4);
(7-3) calculating the gradient value and the gradient direction of each point in the rectangular window, and then sequencing all the points according to the gradient values;
(7-4) judging each point in the sorted point set, eliminating pixel points with gradient amplitudes smaller than a set threshold value, and combining pixel points with similar directions in the rest pixel points by a region growing method;
(7-5) performing rectangle estimation on the discrete point set obtained by the region growing method, including all discrete points in a rectangle box, and calculating the length L, the angle an1 and the area A1 of the rectangle box; rectangular boxes satisfying L1-L2, ani 1-an 1-ana 1 and Arei 1-A1-Area 1 are extracted as line segments to serve as local wire-tripping regions.
8. The detection and classification method based on the chemical fiber yarn spindle and yarn tripwire defect online detection and classification device according to claim 1, characterized in that the specific steps of the steps (1-3-6) are as follows:
(8-1) setting a length prior threshold L2 of a connected domain
Is Ratio;
(8-2) splicing the rectangular frames obtained in the step (7-5) according to the coordinate information of the rectangular frames to obtain a wire tripping mask image corresponding to a wire cake area;
(8-3) carrying out morphological corrosion operation on the tripwire mask image to obtain a corroded mask image;
(8-4) carrying out independent connected domain calculation on the corroded mask image, calculating the length and the area of each connected domain, filling the connected domains with the lengths smaller than a priori threshold L2 into black,
filling black in the connected domain larger than the prior threshold value Ratio to obtain a new mask image;
(8-5) counting the length of each connected domain, the number of the connected domains and the relative position information of the connected domains in the new mask image generated in the step (8-4), taking the length of the connected domains as the tripwire length, taking the number of the connected domains as the tripwire number and taking the relative position information of the connected domains as the position information of the tripwire.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811168402.7A CN109454006B (en) | 2018-10-08 | 2018-10-08 | Detection and classification method based on device for online detection and classification of chemical fiber spindle tripping defects |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811168402.7A CN109454006B (en) | 2018-10-08 | 2018-10-08 | Detection and classification method based on device for online detection and classification of chemical fiber spindle tripping defects |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109454006A CN109454006A (en) | 2019-03-12 |
| CN109454006B true CN109454006B (en) | 2020-06-16 |
Family
ID=65607337
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811168402.7A Active CN109454006B (en) | 2018-10-08 | 2018-10-08 | Detection and classification method based on device for online detection and classification of chemical fiber spindle tripping defects |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109454006B (en) |
Families Citing this family (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110441318A (en) * | 2019-08-22 | 2019-11-12 | 郑州大学 | A kind of chemical fibre spinneret hole defect inspection method based on machine vision |
| CN110827264A (en) * | 2019-11-06 | 2020-02-21 | 四川济通工程试验检测有限公司 | Evaluation system for apparent defects of concrete member |
| CN110838149B (en) * | 2019-11-25 | 2020-10-23 | 创新奇智(广州)科技有限公司 | Camera light source automatic configuration method and system |
| CN111141741B (en) * | 2019-12-09 | 2023-02-10 | 合肥中科迪宏自动化有限公司 | Detection apparatus capable of identifying label defects of printer |
| CN113533359A (en) * | 2020-04-16 | 2021-10-22 | 杭州慧知连科技有限公司 | Visual detection method and device for surface defects of silk ingots |
| CN113533357B (en) * | 2020-04-16 | 2023-09-29 | 杭州慧知连科技有限公司 | Visual inspection method and inspection device for surface defects of silk ingot |
| CN111583202B (en) * | 2020-04-27 | 2023-09-01 | 浙江华睿科技股份有限公司 | Method and device for detecting broken filaments |
| CN111579414B (en) * | 2020-05-21 | 2022-07-26 | 广东精迅里亚特种线材有限公司 | Enameled wire classification control method and system |
| CN111815564B (en) * | 2020-06-09 | 2024-03-29 | 浙江华睿科技股份有限公司 | Method and device for detecting silk ingots and silk ingot sorting system |
| CN111805541B (en) * | 2020-07-08 | 2022-08-30 | 南京航空航天大学 | Deep learning-based traditional Chinese medicine decoction piece cleaning and selecting device and cleaning and selecting method |
| CN112278693B (en) * | 2020-08-26 | 2022-06-03 | 江苏卓高新材料科技有限公司 | Software-based balancing method, system, device, memory, processor and equipment |
| CN112330595B (en) * | 2020-10-13 | 2024-04-02 | 浙江华睿科技股份有限公司 | Method and device for detecting stumbling wire, electronic equipment and storage medium |
| CN112330598B (en) * | 2020-10-14 | 2023-07-25 | 浙江华睿科技股份有限公司 | Method, device and storage medium for detecting stiff yarn defects on chemical fiber surface |
| CN112819780A (en) * | 2021-01-29 | 2021-05-18 | 菲特（天津）检测技术有限公司 | Method and system for detecting surface defects of silk ingots and silk ingot grading system |
| CN112964732A (en) * | 2021-02-04 | 2021-06-15 | 科大智能物联技术有限公司 | Spinning cake defect visual detection system and method based on deep learning |
| CN112686227B (en) * | 2021-03-12 | 2021-07-06 | 泰瑞数创科技(北京)有限公司 | Product quality inspection method and device based on augmented reality and man-machine comprehensive detection |
| CN113096073B (en) * | 2021-03-19 | 2022-10-18 | 浙江华睿科技股份有限公司 | Method and device for detecting surface flatness of chemical fiber spindle |
| CN113724258B (en) * | 2021-11-02 | 2022-02-08 | 山东中都机器有限公司 | Conveyor belt tearing detection method and system based on image processing |
| CN114820625B (en) * | 2022-06-30 | 2022-09-09 | 爱普车辆股份有限公司 | Automobile top block defect detection method |
| CN115984282B (en) * | 2023-03-21 | 2023-06-16 | 菲特（天津）检测技术有限公司 | Spandex product detection method, device, equipment and storage medium |
| CN116452598B (en) * | 2023-06-20 | 2023-08-29 | 曼德惟尔(山东)智能制造有限公司 | Axle production quality rapid detection method and system based on computer vision |
| CN116823825B (en) * | 2023-08-29 | 2023-12-05 | 山东海德尼克液压科技有限公司 | Intelligent identification method for casting defects of valve fittings |
| CN116990229B (en) * | 2023-09-26 | 2023-12-08 | 南通赛可特电子有限公司 | Defect detection method and system for copper plating layer surface of product |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104778692B (en) * | 2015-04-09 | 2017-08-04 | 中原工学院 | A kind of fabric defect detection method optimized based on rarefaction representation coefficient |
| CN204718967U (en) * | 2015-06-11 | 2015-10-21 | 龚厚志 | The accurate disposal system of image in a kind of chemical fiber paper tube outward appearance testing process |
| CN106408557A (en) * | 2016-06-30 | 2017-02-15 | 江汉大学 | Viscose long filament and broken filament detection method based on image processing technology |
| CN106841209B (en) * | 2016-12-28 | 2019-05-24 | 杭州慧芯智识科技有限公司 | One kind being based on big data self study chemical fiber wire ingot intelligence appearance detection system and method |
| CN108508035B (en) * | 2018-05-10 | 2023-11-24 | 杭州慧知连科技有限公司 | Intelligent detection equipment for appearance defects of chemical fiber silk ingots |
| CN108362701A (en) * | 2018-05-18 | 2018-08-03 | 西安获德图像技术有限公司 | A kind of chemical fiber spinning cake appearance automatic checkout system |
-
2018
- 2018-10-08 CN CN201811168402.7A patent/CN109454006B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN109454006A (en) | 2019-03-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109454006B (en) | Detection and classification method based on device for online detection and classification of chemical fiber spindle tripping defects | |
| CN109550712B (en) | Chemical fiber filament tail fiber appearance defect detection system and method | |
| CN109255787B (en) | System and method for detecting scratch of silk ingot based on deep learning and image processing technology | |
| CN109490316B (en) | Surface defect detection algorithm based on machine vision | |
| CN109724984B (en) | Defect detection and identification device and method based on deep learning algorithm | |
| CN104112269B (en) | A kind of solar battery laser groove parameter detection method and system based on machine vision | |
| CN105388162B (en) | Raw material silicon chip surface scratch detection method based on machine vision | |
| CN108765402B (en) | Non-woven fabric defect detection and classification method | |
| CN108760747A (en) | A kind of 3D printing model surface defect visible detection method | |
| JPH0434068A (en) | Inspection of top cleanliness | |
| CN117635565B (en) | Semiconductor surface defect detection system based on image recognition | |
| CN115222709B (en) | Zipper cloth belt defect detection method based on artificial intelligence | |
| CN205538710U (en) | Inductance quality automatic check out system based on machine vision | |
| CN102175692A (en) | System and method for detecting defects of fabric gray cloth quickly | |
| CN113222938A (en) | Chip defect detection method and system and computer readable storage medium | |
| CN109345524A (en) | A kind of bearing open defect detection system of view-based access control model | |
| CN110349125A (en) | A kind of LED chip open defect detection method and system based on machine vision | |
| CN110096980A (en) | Character machining identifying system | |
| CN110335233A (en) | Express-way guard-rail plates defect detecting system and method based on image processing techniques | |
| CN116434206A (en) | Cotton quality character detection method based on machine vision technology | |
| CN110813795B (en) | Device and method for detecting breakage of chemical fiber spindle paper tube on line | |
| Zhang et al. | Defect detection of mobile phone screen based on improved difference image method | |
| CN115240144B (en) | Method and system for intelligently identifying flaws in spinning twisting | |
| CN115672778A (en) | Intelligent visual recognition system | |
| CN205711109U (en) | A kind of CCD electronic imaging foreign fiber piece-rate system |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |