CN112164048B - Magnetic shoe surface defect automatic detection method and device based on deep learning - Google Patents
Magnetic shoe surface defect automatic detection method and device based on deep learning Download PDFInfo
- Publication number
- CN112164048B CN112164048B CN202011025134.0A CN202011025134A CN112164048B CN 112164048 B CN112164048 B CN 112164048B CN 202011025134 A CN202011025134 A CN 202011025134A CN 112164048 B CN112164048 B CN 112164048B
- Authority
- CN
- China
- Prior art keywords
- image
- magnetic
- surface defects
- defect
- automatic detection
- 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
-
- 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
- G06T7/0008—Industrial image inspection checking presence/absence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Abstract
The invention discloses a method and a device for automatically detecting surface defects of magnetic shoes based on deep learning, and belongs to the field of automatic detection of surface defects of objects. The method of the invention fully utilizes the characteristics of the surface defects of the magnetic tiles, correspondingly calculates four saliency maps of the image to be detected, fuses the four saliency maps obtained by calculation with the original image, inputs the fused image into a U-net network to realize pixel level segmentation, can make pixel level judgment on the size and the position of the defects, and effectively improves the accuracy of the detection result; the method does not need preprocessing processes such as correction, image enhancement, denoising, feature extraction and the like on the input image, and has good robustness on image deformation, environmental illumination change and angle change. The device is simple in configuration, can be embedded into production without greatly changing the existing production line, and reduces resource waste in the implementation process of the scheme.
Description
Technical Field
The invention belongs to the field of automatic detection of surface defects of objects, and particularly relates to a method and a device for automatically detecting surface defects of magnetic shoes based on deep learning.
Background
In recent years, in order to improve the competitiveness of products, each magnetic material production enterprise urgently needs an advanced full-automatic wet press, an automatic grinding line and an automatic magnetic shoe detection line, so that production equipment can reach an international advanced level to adapt to international competition. The generation of magnetic tiles is a labor intensive industry worldwide and is done manually. The manual detection has the advantages of good flexibility and high adaptability to variety change, but the detection for a long time is very difficult and extremely unstable; meanwhile, along with the annual rise of labor cost, the automation level of the magnetic tile production line is urgently needed to be improved.
However, some existing workpiece defect detection technologies based on machine vision cannot achieve high accuracy or have complex device structures when facing different defect types of magnetic shoes; for example, CN110702696a discloses a magnetic shoe surface defect detection device, which comprises a magnetic shoe, a telecentric light source, a prism group, a cylindrical lens group, a bar-shaped light source group, a planar coaxial light source, a telecentric lens and a camera. The detection device of the method is complex, and is difficult to be embedded into actual production on the basis of not modifying the existing production line, and the modification of the production line can bring waste of resources and time. For another example, CN110827263A discloses a magnetic tile surface defect detection system based on a visual recognition technology, and is used for detecting whether a defect exists on the surface of a magnetic tile. Although the method adopts a deep learning mode, the method can only judge whether the sample has defects, and cannot judge the size and the position of the defects at a pixel level.
Disclosure of Invention
Aiming at the defects or improvement requirements in the prior art, the invention provides a method and a device for automatically detecting the surface defects of the magnetic tiles based on deep learning, and aims to provide the method and the device for automatically detecting the surface defects of the magnetic tiles, which have a simple structure and high accuracy of detection results.
In order to achieve the above object, according to an aspect of the present invention, there is provided a method for automatically detecting surface defects of a magnetic shoe based on deep learning, including:
s1, constructing an automatic detection model of the surface defects of the magnetic shoe; the automatic detection model for the surface defects of the magnetic tiles comprises a saliency map calculation module and a saliency map detection moduleA sign fusion module and a U-net network; wherein, the saliency map calculation module is used for calculating the AC saliency map S of the input image AC ST significant map S ST PHOT significant map S PHOT And BMS saliency map S BMS ；S AC For capturing rarity, S, of defect color ST For characterizing edges and corners, S PHOT Eliminating the interference of background texture; s BMS For simulating a human visual attention mechanism to detect salient objects; the feature fusion module is used for fusing the input original image and the saliency map obtained by calculation through the convolution layer; the U-net network is used for carrying out pixel level segmentation on the feature map obtained by fusion to obtain the position and the size of the defect;
s2, performing iterative training on the automatic detection model for the surface defects of the magnetic shoes to obtain a trained automatic detection model for the surface defects of the magnetic shoes;
and S3, inputting the magnetic shoe surface image to be detected into the trained automatic magnetic shoe surface defect detection model to obtain the position and the size of the magnetic shoe surface image defect to be detected.
Further, the feature fusion module employs a 1 × 1 convolution kernel with a depth of 5.
Further, the input image sizes are all 512 × 512 × 1.
Further, the U-net network comprises three parts of down-sampling, up-sampling and hopping connection.
Further, the downsampling and upsampling portions both use the Relu activation function.
According to another aspect of the present invention, there is provided an automatic detection apparatus for surface defects of magnetic tiles based on deep learning, comprising: the system comprises a conveyor belt, a limiter, a stand column, a cross beam, an RGB camera, a network data exchange router and a processor;
the upright post is fixed on the ground through a bolt, and a row of circular through holes are formed in the upright post and used for fixing the limiter and the cross beam; the RGB camera is arranged on the cross beam and fixed above the conveyor belt; the network data exchange router is connected with the RGB camera and the processor;
placing a sample image to be detected in the center of the conveyor belt;
the RGB camera is used for collecting a sample image to be detected currently on the transmission belt;
the network data exchange router is used for transmitting the image acquired by the RGB camera to the processor;
the processor is internally provided with a trained automatic detection model for the surface defects of the magnetic shoes, and is used for preprocessing the image of the sample to be detected, calculating four saliency maps of an original image, fusing the four saliency maps and the original image, and inputting the fused image into a U-net network to obtain the position and the size of the defects in the image of the sample to be detected.
Further, the automatic detection model for the surface defects of the magnetic tiles comprises a saliency map calculation module, a feature fusion module and a U-net network; wherein, the saliency map calculation module is used for calculating the AC saliency map S of the input image AC ST significant map S ST PHOT significant map S PHOT And BMS saliency map S BMS ；S AC For capturing rarity, S, of defect color ST For characterizing edges and corners, S PHOT Eliminating the interference of background texture; s BMS For simulating a human visual attention mechanism to detect salient objects; the feature fusion module is used for fusing the input original image and the saliency map obtained by calculation through the convolution layer; and the U-net network is used for carrying out pixel level segmentation on the feature map obtained by fusion to obtain the position and the size of the defect.
In general, the above technical solutions contemplated by the present invention can achieve the following advantageous effects compared to the prior art.
(1) The method of the invention fully utilizes the characteristics of the surface defects of the magnetic tiles, correspondingly calculates four saliency maps of the image to be detected, fuses the four saliency maps obtained by calculation with the original image, inputs the fused image into the U-net network to realize pixel level segmentation, can make pixel level judgment on the size and the position of the defects, and effectively improves the accuracy of the detection result.
(2) The method does not need preprocessing processes such as correction, image enhancement, denoising, feature extraction and the like on the input image, and has good robustness on image deformation, environmental illumination change and angle change.
(3) The device is simple in configuration, can be embedded into production without greatly changing the existing production line, and reduces resource waste in the implementation process of the scheme.
(4) The invention has wide range of detection objects, and similar processes requiring traditional visual detection can be simplified by adopting the invention, thus realizing automatic detection and rapid detection.
Drawings
Fig. 1 is a structural diagram of a magnetic tile surface defect detection method of a U-net network according to an embodiment of the present invention.
FIG. 2 is a schematic structural diagram of an apparatus for automatically detecting surface defects of magnetic shoes according to an embodiment of the present invention;
the device comprises a magnetic tile sample to be detected 101, a conveyor belt 102, a stopper 103, a bolt 104, a column 105, a beam 106, an RGB camera 107, a data line 108, a network data exchange router 109 and a processor 110.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the respective embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
The invention provides a method and a device for automatically detecting surface defects of magnetic tiles based on deep learning. The automation level of the magnetic shoe surface defect detection can be effectively improved, and the detection speed and the accuracy of the detection result are improved.
The traditional visual detection of the surface defects of the magnetic tiles is often influenced by human factors, and particularly, the human factors have more obvious influence on detection results when people are in a visual fatigue state. Deep learning realizes complex function approximation and input data representation by learning a deep nonlinear network structure, and shows the learning capability of strong essential characteristics of a data set. The full convolution neural network (FCN), as a typical deep learning method, can classify images at a pixel level, thereby solving the problem of image segmentation at a semantic level. Unlike the classic CNN which uses a full link layer to obtain a fixed-length feature vector for classification (full link layer + softmax output) after the convolutional layer, the FCN can accept an input image of any size, and uses the deconvolution layer to upsample the feature map of the last convolutional layer to restore it to the same size as the input image, thereby generating a prediction for each pixel while retaining spatial information in the original input image, and finally performing pixel-by-pixel classification on the upsampled feature map. U-net is an important extension of FCN, originally designed for biomedical cellular image segmentation. This design not only avoids loss of critical information during sharing, but also enables the model to perform pixel-to-pixel and end-to-end learning. In particular, the convolutional layers of the U-net encoder part are directly copied and cropped to obtain upsampled layers of the same size, skipping fixed layer connections helps to repair details lost in the pooling process. Similar architectures have been applied to many advanced detection models to obtain more accurate detection results.
The magnetic shoe surface defect automatic detection device based on deep learning can acquire images of a sample to be detected of a magnetic shoe in real time, the images of the sample are transmitted to a processor, pixel-level segmentation is carried out on the images by using a depth model, and the sample with defects is automatically detected, so that adverse effects caused by human factors are avoided. The industrial high-definition RGB camera is used for collecting the image of the sample to be detected of the magnetic shoe, then the image is connected to the network data exchange router through the data line, the image data is sent to the network data exchange router, the network data exchange router is connected to the processor through the data line, and then the image data is sent to the processor for detection. After the image is input into the processor, firstly four saliency maps of an original image are calculated, the four saliency maps and the original image are fused, the fused image is input into the U-net network, feature learning is carried out on the fused image by using the U-net network to obtain a deep learning model, and then defect detection is carried out on a sample to be detected by using the deep learning model to obtain a detection result.
Specifically, an embodiment of the present invention provides an automatic detection method for a magnetic tile surface defect based on deep learning, including:
s1, constructing an automatic detection model of the surface defects of the magnetic shoes; the automatic detection model for the surface defects of the magnetic tiles comprises a saliency map calculation module, a feature fusion module and a U-net network; wherein, the saliency map calculation module is used for calculating the AC saliency map S of the input image AC ST significant map S ST PHOT significant map S PHOT And BMS saliency map S BMS ；S AC For capturing rarity, S, of defect color ST For characterizing edges and corners, S PHOT Eliminating the interference of background texture; s BMS For simulating a human visual attention mechanism to detect salient objects; the feature fusion module is used for fusing the input original image and the saliency map obtained by calculation through the convolution layer; the U-net network is used for carrying out pixel level segmentation on the feature map obtained by fusion to obtain the position and the size of the defect;
as shown in fig. 1, the four saliency maps of the original image are chosen as follows:
the surface defects of the magnetic shoe have the following four remarkable characteristics:
first, the color of the defect location is different from the surrounding environment, and its gray value is often low. Aiming at the salient feature, the invention selects a full-resolution algorithm (AC algorithm) to calculate an AC salient map S of the full-resolution algorithm AC ；
Second, the corner and edge response of the defect is strong. Therefore, the invention adopts a Structure Tensor (ST) model to highlight the edge, and the ST model is marked as an ST saliency map S ST ；
Third, the texture of the background region of the image is relatively regular. Aiming at the remarkable characteristicSymbolizing that a Phase Only Transform (PHOT) model is adopted to calculate the PHOT saliency map S PHOT 。
Fourth, defects such as vent cracks can be easily detected using human visual attention. Aiming at the salient feature, the invention adopts a Boolean Map-based salient model (BMS) to calculate the BMS salient Map S of the BMS salient model BMS 。
In particular, it is generally desirable to fuse several saliency maps in a weighted summation manner, but artificially setting weights has uncertainty on the learned effect. Therefore, the invention designs a convolution layer, and learns the optimal weight by utilizing the back propagation of the neural network. The input of the convolution layer is four saliency maps and one original image, the image sizes are 512 × 512 × 1, and a 1 × 1 convolution kernel with a depth of 5 is adopted. And inputting the fused image into a U-net network for training.
Specifically, the input layer of the U-net network is 512 multiplied by 1; the U-net network mainly comprises a down-sampling part, an up-sampling part and a jump connection part; the down-sampling part comprises a plurality of conventional 3 x 3 convolutions and a 2 x 2 maximum value pooling, the up-sampling part comprises a plurality of 2 x 2 deconvolution and a conventional 3 x 3 convolution, and the down-sampling part and the up-sampling part both adopt Relu activating functions.
Specifically, the Relu activation function is expressed as follows:
f(x)＝max(0，x)。
s2, performing iterative training on the automatic detection model for the surface defects of the magnetic shoes to obtain a trained automatic detection model for the surface defects of the magnetic shoes;
and S3, inputting the magnetic shoe surface image to be detected into the trained automatic magnetic shoe surface defect detection model to obtain the position and the size of the magnetic shoe surface image defect to be detected.
As shown in fig. 2, an embodiment of the present invention further provides an automatic detection apparatus for a surface defect of a magnetic tile based on deep learning, including: the system comprises a conveyor belt 102, a stopper 103, a bolt 104, a column 105, a beam 106, an RGB camera 107, a data line 108, a network data exchange router 109 and a processor 110;
the overall position relation of the devices is as follows: the stopper 103 is fixed on the column 105 by bolts, the column 105 is fixed on the ground by bolts, the beam 106 is fixed on the rigid column 105 by bolts, and the RGB camera 107 is mounted on the rigid beam 106 by bolts and fixed above the conveyor belt 102. The network data exchange router 109 may connect the RGB camera 107 and the processor 110 through a data line, and the sample image data to be detected may be transmitted to the processor 110 through the network data exchange router 109 through the data line.
The limiting device 103 and the cross beam 106 are both provided with square through holes and bolt holes, the cross beam 106 is provided with a wire groove for arranging data wires, and a threaded hole for fixing the RGB camera 107 is formed below the cross beam 106; a column of circular through holes is formed on the column 105 for fixing the stopper 103 and the beam 106.
Specifically, as shown in fig. 2, the magnetic tile sample 101 to be detected is placed in the center of the conveyor belt 102 and does not exceed the edge position of the conveyor belt, and the conveyor belt 102 conveys the magnetic tile sample to be detected below the camera 107 at a certain speed. The upright post, the limiter, the cross beam and the RGB camera form a whole; the RGB camera 107 is positioned above the sample to be detected, the position of the RGB camera 107 along the direction of the chute is adjusted by adjusting the bolt 104 at the bottom of the upright column, the height of the RGB camera 107 is adjusted by adjusting the bolt of the cross beam 106, and the angle of the RGB camera 107 is adjusted, so that the imaging plane of the camera is parallel to the conveyor belt, the visual field of the camera can be ensured to cover and support the sample to be detected currently on the conveyor belt 102, and a complete sample image is acquired; the RGB camera 107 acquires a current sample image to be detected, transmits image data to the processor 110 via the network data exchange router 109, embeds a trained automatic detection model for magnetic tile surface defects in the processor 110, and performs sample analysis based on deep learning on the received image data, thereby obtaining a detection result.
The RGB camera 107 may be a berlin S908 industrial wide-angle camera.
The following detailed description of the embodiments of the invention refers to the accompanying drawings in which:
first, installation of the device: fixing the pillar 105 to the floor on one side of the conveyor belt 102; the stopper 103 is fixed at a proper height of the upright column by a bolt 104, and the height that the camera 107 does not touch the conveyor belt is taken as the lower limit; the cross beam 106 is fixed at a proper height through bolts; the camera 107 is fixed in place by bolts as per fig. 1; and the camera, the network data exchange router and the processor are connected by using a data line.
Secondly, equipment debugging: before the detection starts, the data transmission link needs to be checked, and the position of the camera 107 needs to be adjusted, so that the camera is ensured to acquire a smooth image; adjusting the spatial position of the camera by adjusting bolts at the bottoms of the upright posts, the fixed part of the cross beam and the fixed part of the camera; the angle of the camera is adjusted through the camera self-mechanism.
Then, the detection starts: the method comprises the steps that magnetic tile samples 101 to be detected are placed at the center of a conveyor belt 102 at equal intervals, the conveyor belt conveys the magnetic tile samples 101 to be detected to the position below a camera 107 of the conveyor belt according to a certain speed, the industrial high-definition camera 107 shoots images of the samples to be detected, the images are transmitted to a processor 110 through a data line 108 and a network data exchange router 109, the images are processed in a deep learning network model after being input into the processor, the trained network autonomously detects whether the samples have defects or not, areas where the surface defects of the samples are located are marked out, and finally automatic detection of the surface defects of the magnetic tiles is achieved. The data transmission mode is not limited to the wired transmission mode, and may also be a wireless data transmission mode, specifically, which data transmission mode is adopted, and the embodiment of the present invention is not limited uniquely.
It will be understood by those skilled in the art that the foregoing is only an exemplary embodiment of the present invention, and is not intended to limit the invention to the particular forms disclosed, since various modifications, substitutions and improvements within the spirit and scope of the invention are possible and within the scope of the appended claims.
Claims (6)
1. A magnetic shoe surface defect automatic detection method based on deep learning is characterized by comprising the following steps:
s1, constructing an automatic detection model for surface defects of magnetic shoes(ii) a The automatic detection model for the surface defects of the magnetic tiles comprises a saliency map calculation module, a feature fusion module and a U-net network; wherein, the saliency map calculation module is used for calculating the AC saliency map S of the input image AC ST significant map S ST PHOT significant map S PHOT And BMS saliency map S BMS ；S AC Rarity, S for highlighting defect color ST For highlighting edges and corners of defects, S PHOT The method is used for eliminating the interference of background textures on the defect area; s. the BMS The system is used for simulating a human visual attention mechanism to highlight the defect area; the feature fusion module is used for fusing the input original image and the four calculated saliency maps through the convolution layer; the U-net network is used for carrying out pixel level segmentation on the feature map obtained by fusion to obtain the position and the size of the defect;
s2, performing iterative training on the automatic detection model for the surface defects of the magnetic shoes to obtain a trained automatic detection model for the surface defects of the magnetic shoes;
and S3, inputting the magnetic shoe surface image to be detected into the trained automatic magnetic shoe surface defect detection model to obtain the position and the size of the magnetic shoe surface image defect to be detected.
2. The method for automatically detecting the surface defects of the magnetic tiles based on the deep learning as claimed in claim 1, wherein the feature fusion module adopts a convolution kernel with the depth of 5 and the 1 x 1.
3. The method as claimed in claim 1, wherein the input image size is 512 x 1.
4. The method for automatically detecting the surface defects of the magnetic tiles based on the deep learning of claim 1, wherein the U-net network comprises down-sampling, up-sampling and jump connection.
5. The method for automatically detecting the surface defects of the magnetic tiles based on the deep learning as claimed in claim 4, wherein the downsampling part and the upsampling part both adopt Relu activation functions.
6. The utility model provides a magnetic shoe surface defect automatic checkout device based on deep learning which characterized in that includes: the system comprises a conveyor belt (102), a stopper (103), a stand column (105), a beam (106), an RGB camera (107), a network data exchange router (109) and a processor (110);
the upright post (105) is fixed on the ground through bolts, and a row of circular through holes are formed in the upright post and used for fixing the limiter (103) and the cross beam (106); the RGB camera (107) is arranged on the cross beam (106) and fixed above the conveyor belt (102); the network data exchange router (109) is connected with the RGB camera (107) and the processor (110);
the sample image (101) to be detected is arranged in the center of the conveyor belt (102);
the RGB camera (107) is used for collecting a current sample image to be detected on the conveyor belt (102);
the network data exchange router (109) is used for transmitting the image collected by the RGB camera (107) to the processor (110);
a trained automatic detection model for the surface defects of the magnetic tiles is arranged in the processor (110) and used for preprocessing the sample image to be detected, calculating four saliency maps of an original image, fusing the four saliency maps and the original image, and inputting the fused image into a U-net network to obtain the position and the size of the defects in the sample image to be detected; the automatic detection model for the surface defects of the magnetic tiles comprises a saliency map calculation module, a feature fusion module and a U-net network; wherein, the saliency map calculation module is used for calculating the AC saliency map S of the input image AC ST significant map S ST PHOT significant map S PHOT And BMS saliency map S BMS ；S AC Rarity, S for highlighting defect color ST For highlighting edges and corners of defects, S PHOT The method is used for eliminating the interference of background textures on the defect area; s. the BMS The system is used for simulating a human visual attention mechanism to highlight the defect area; a feature fusion module for scrolling the input original image and the four calculated saliency mapsLaminating and fusing; and the U-net network is used for carrying out pixel level segmentation on the feature map obtained by fusion to obtain the position and the size of the defect.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011025134.0A CN112164048B (en) | 2020-09-25 | 2020-09-25 | Magnetic shoe surface defect automatic detection method and device based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011025134.0A CN112164048B (en) | 2020-09-25 | 2020-09-25 | Magnetic shoe surface defect automatic detection method and device based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112164048A CN112164048A (en) | 2021-01-01 |
| CN112164048B true CN112164048B (en) | 2023-03-10 |
Family
ID=73864074
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011025134.0A Active CN112164048B (en) | 2020-09-25 | 2020-09-25 | Magnetic shoe surface defect automatic detection method and device based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112164048B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112801962B (en) * | 2021-01-19 | 2022-09-16 | 上海大学 | Semi-supervised industrial product flaw detection method and system based on positive sample learning |
| CN113177921A (en) * | 2021-04-30 | 2021-07-27 | 佛山市南海区广工大数控装备协同创新研究院 | Magnetic shoe surface defect detection method based on neural network |
| CN113298765B (en) * | 2021-05-19 | 2023-11-07 | 广东工业大学 | Glass panel surface defect detection method based on image processing and deep learning |
| CN114581362B (en) * | 2021-07-22 | 2023-11-07 | 正泰集团研发中心(上海)有限公司 | Photovoltaic module defect detection method and device, electronic equipment and readable storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107437246A (en) * | 2017-07-05 | 2017-12-05 | 浙江大学 | A kind of common conspicuousness detection method based on end-to-end full convolutional neural networks |
| CN108230324A (en) * | 2018-01-31 | 2018-06-29 | 浙江理工大学 | Magnetic shoe surface microdefect visible detection method |
| CN110599470A (en) * | 2019-08-30 | 2019-12-20 | 武汉科技大学 | Magnetic shoe surface defect detection system and method |
| CN110827263A (en) * | 2019-11-06 | 2020-02-21 | 创新奇智(南京)科技有限公司 | Magnetic shoe surface defect detection system and detection method based on visual identification technology |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10699151B2 (en) * | 2016-06-03 | 2020-06-30 | Miovision Technologies Incorporated | System and method for performing saliency detection using deep active contours |
-
2020
- 2020-09-25 CN CN202011025134.0A patent/CN112164048B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107437246A (en) * | 2017-07-05 | 2017-12-05 | 浙江大学 | A kind of common conspicuousness detection method based on end-to-end full convolutional neural networks |
| CN108230324A (en) * | 2018-01-31 | 2018-06-29 | 浙江理工大学 | Magnetic shoe surface microdefect visible detection method |
| CN110599470A (en) * | 2019-08-30 | 2019-12-20 | 武汉科技大学 | Magnetic shoe surface defect detection system and method |
| CN110827263A (en) * | 2019-11-06 | 2020-02-21 | 创新奇智(南京)科技有限公司 | Magnetic shoe surface defect detection system and detection method based on visual identification technology |
Non-Patent Citations (3)
| Title |
|---|
| Luofeng Xie.et..FFCNN:A Deep Neural Network for Surface Defect Detection of Magnetic Tile.《IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS》.2020,第68卷(第4期),第3506-3516页. * |
| Yunlong Dong.et..Principled reward shaping for reinforcement learning via lyapunov stability theory.《Neurocomputing》.2020,第393卷第83-90页. * |
| 孙海涛.基于机器视觉的磁瓦表面缺陷检测系统.《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》.2017,(第3期),第I138-5035页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112164048A (en) | 2021-01-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112164048B (en) | Magnetic shoe surface defect automatic detection method and device based on deep learning | |
| CN106548182B (en) | Pavement crack detection method and device based on deep learning and main cause analysis | |
| CN105279372B (en) | A kind of method and apparatus of determining depth of building | |
| CN111507958B (en) | Target detection method, training method of detection model and electronic equipment | |
| WO2020137222A1 (en) | Defect inspecting device, defect inspecting method, and program for same | |
| CN109034184B (en) | Grading ring detection and identification method based on deep learning | |
| CN114897864B (en) | Workpiece detection and defect judgment method based on digital-analog information | |
| CN111507976B (en) | Defect detection method and system based on multi-angle imaging | |
| JP2018190329A (en) | Image processing apparatus, image processing program, and image processing system | |
| CN110596120A (en) | Glass boundary defect detection method, device, terminal and storage medium | |
| CN112330593A (en) | Building surface crack detection method based on deep learning network | |
| CN111259710B (en) | Parking space structure detection model training method adopting parking space frame lines and end points | |
| CN112304960B (en) | High-resolution image object surface defect detection method based on deep learning | |
| CN111597941B (en) | Target detection method for dam defect image | |
| CN115156093A (en) | Battery shell defect detection method, system and device | |
| CN115830004A (en) | Surface defect detection method, device, computer equipment and storage medium | |
| CN115018846A (en) | AI intelligent camera-based multi-target crack defect detection method and device | |
| CN112200790A (en) | Cloth defect detection method, device and medium | |
| CN115775236A (en) | Surface tiny defect visual detection method and system based on multi-scale feature fusion | |
| CN116071315A (en) | Product visual defect detection method and system based on machine vision | |
| CN113706496B (en) | Aircraft structure crack detection method based on deep learning model | |
| CN110717910B (en) | CT image target detection method based on convolutional neural network and CT scanner | |
| CN113781512A (en) | Image boundary identification method, device, equipment, system and storage medium | |
| Xu et al. | A Deep Neural Network-Based Intelligent Detection Model for Manufacturing Defects of Automobile Parts | |
| Shah et al. | Sign Detection Vision Based Mobile Robot Platform |
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 |