CN110648316B - Steel coil end face edge detection method based on deep learning - Google Patents
Steel coil end face edge detection method based on deep learning Download PDFInfo
- Publication number
- CN110648316B CN110648316B CN201910844889.4A CN201910844889A CN110648316B CN 110648316 B CN110648316 B CN 110648316B CN 201910844889 A CN201910844889 A CN 201910844889A CN 110648316 B CN110648316 B CN 110648316B
- Authority
- CN
- China
- Prior art keywords
- scale
- convolution
- feature
- feature extraction
- edge 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a steel coil end face edge detection method based on deep learning, which comprises the steps of improving a deep convolutional neural network structure, and respectively improving a multi-scale feature extraction part and a multi-scale feature fusion part, wherein the multi-scale feature extraction part is composed of five feature extraction modules with different scales, the multi-scale feature fusion part is mainly composed of a convolutional layer and a reverse convolutional layer, image acquisition is carried out, an acquired image is sent into the deep convolutional neural network for edge detection, and then an edge detection result is output; the deep convolution neural network is improved, automatically learns the characteristics from the data, can learn the edge of neglected numbers, only shows the edge of the steel coil, and has better robustness and good detection effect.
Description
Technical Field
The invention relates to the technical field of industrial vision, in particular to a steel coil end face edge detection method based on deep learning.
Background
The steel coil end face has extremely strong regular texture edge characteristics, and in order to automatically detect the defect condition of the steel coil end face, the edge of the steel coil end face needs to be detected in the first step, so that auxiliary information is provided for subsequent defect detection.
In the prior art, the edge detection algorithm based on the traditional image processing is usually based on the angle of calculating the pixel gradient of an image, various different edge detection operators are adopted, and the sliding window calculation is carried out on the image in a convolution filtering mode, wherein the most common edge detection algorithm is the canny edge detection algorithm. However, when the method is applied to the field of edge detection of the end face of the steel coil, the edge image obtained based on the traditional image processing algorithm often has a plurality of noise points, such as high light and shadow generated by illumination, and uneven parts generated by oxidation on the surface of the steel coil can greatly interfere with the edge image detected by the traditional algorithm, and the loss function of the traditional edge detection adopts a pixel-by-pixel binary cross entropy loss function, and then the pixel-by-pixel loss values are summed to obtain a final loss value, namely the final loss value is obtained, namely the final loss value
Because the edge lines occupy few pixels and the background pixels are very many, the two types of class imbalance can cause that the network training process is easy to fall into a local extremum, and the accuracy of the calculation result is low.
Disclosure of Invention
The invention aims to overcome the problems in the prior art and provides a steel coil end face edge detection method based on deep learning.
In order to achieve the technical purpose and achieve the technical effect, the invention is realized by the following technical scheme:
a steel coil end face edge detection method based on deep learning comprises the following steps:
s1, improving the structure of the deep convolutional neural network, and respectively improving the multi-scale feature extraction part and the multi-scale feature fusion part;
s101, a multi-scale feature extraction part consists of five feature extraction modules with different scales, each module comprises two to three convolution layers with convolution kernels of 3 x 3 and a maximum pooling layer with kernels of 2 x 2 and step length of 2, the first scale feature extraction module comprises two convolution layers, the number of convolution filters is 64, the second scale feature extraction module comprises two convolution layers, the number of convolution filters is 128, the subsequent three scale feature extraction modules all comprise three convolution layers, the number of convolution filters is sequentially multiplied by 2 along with scale change, and the scale of a feature map is halved after each pooling layer is passed, so that 5 feature extraction modules output feature maps with 5 different scales;
s102, a multi-scale feature fusion part mainly comprises a convolution layer and an anti-convolution layer, for each scale of feature graph, firstly, the feature graph passes through a 1 x 1 convolution layer, then passes through an anti-convolution layer, the effect of the anti-convolution layer is mainly to perform up-sampling on the feature graph, so that the feature graphs of different scales can be finally changed into feature graphs with the same size as an input image, after 5 feature graphs with the same size are obtained, the 5 feature graphs are spliced together, and then pass through the 1 x 1 convolution layer, the function of the convolution layer is to automatically learn the weights of the feature graphs of different scales, and finally, the 5 feature graphs are subjected to weighted summation to obtain the output of a network;
and S2, collecting the image, sending the collected image into a deep convolution neural network for edge detection, and outputting an edge detection result.
Further, the edge detection step is as follows:
inputting a training sample, carrying out multi-scale feature extraction on the training sample to obtain a plurality of scale feature maps, fusing the plurality of scale feature maps by a multi-scale feature fusion part, and then calculating a category loss function, wherein the category loss function calculation formula is as follows:
where β is the number of background pixels divided by the total number of pixels.
The benefit effects of the invention are:
1. the deep convolutional neural network detects edges by automatically learning features from data and compared with a traditional algorithm in which a gradient operator is manually designed, the deep convolutional neural network has better robustness, and has better detection effects on highlight and shadow generated by illumination and uneven parts generated by oxidation on the surface of a steel coil.
2. The edge image detected by the method is more consistent with semantics, the traditional algorithm detects the edge according to gradient change and generates response to the place where the pixel in the image suddenly changes, however, the deep convolution neural network can consider whether a certain point belongs to the edge according to the surrounding image information, for example, the situation that the end face of the steel coil is painted with numbers, the traditional algorithm cannot distinguish the digital edge from the edge of the steel coil, and the deep convolution neural network can learn the edge which ignores the numbers and only represents the edge of the steel coil.
3. The invention can fully consider the edge characteristics under different scales, for example, when the scale is larger, the local edge characteristics are more obvious, and when the scale is smaller, the global edge has more integrity and is easier to be identified as a whole by the network, and the fusion of the multi-scale characteristic graphs utilizes the advantages of the characteristic graphs of different scales, thereby finally obtaining better edge detection results.
4. By improving the loss function, if only the original binary cross entropy loss function is used, the network is easy to fall into a local extreme point in the training process, for example, the network may tend to output all non-edges, because the proportion of edge pixels to total pixels in a sample is extremely small, the network output all non-edges can obtain a smaller loss value, and the problem does not exist after the class-balanced loss function is adopted.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic overall flow diagram of the present invention;
FIG. 2 is a schematic diagram illustrating the edge detection principle of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1-2, the invention relates to a steel coil end face edge detection method based on deep learning, which comprises the following steps:
s1, improving the structure of the deep convolutional neural network, and respectively improving the multi-scale feature extraction part and the multi-scale feature fusion part;
s101, a multi-scale feature extraction part consists of five feature extraction modules with different scales, each module comprises two to three convolution layers with convolution kernels of 3 x 3 and a maximum pooling layer with kernels of 2 x 2 and step length of 2, the first scale feature extraction module comprises two convolution layers, the number of convolution filters is 64, the second scale feature extraction module comprises two convolution layers, the number of convolution filters is 128, the subsequent three scale feature extraction modules all comprise three convolution layers, the number of convolution filters is sequentially multiplied by 2 along with scale change, and the scale of a feature map is halved after each pooling layer is passed, so that 5 feature extraction modules output feature maps with 5 different scales;
s102, a multi-scale feature fusion part mainly comprises a convolution layer and an anti-convolution layer, for each scale of feature graph, firstly, the feature graph passes through a 1 x 1 convolution layer, then passes through an anti-convolution layer, the effect of the anti-convolution layer is mainly to perform up-sampling on the feature graph, so that the feature graphs of different scales can be finally changed into feature graphs with the same size as an input image, after 5 feature graphs with the same size are obtained, the 5 feature graphs are spliced together, and then pass through the 1 x 1 convolution layer, the function of the convolution layer is to automatically learn the weights of the feature graphs of different scales, and finally, the 5 feature graphs are subjected to weighted summation to obtain the output of a network;
and S2, collecting the image, sending the collected image into a deep convolution neural network for edge detection, and outputting an edge detection result.
The edge detection method comprises the following steps:
inputting a training sample, carrying out multi-scale feature extraction on the training sample to obtain a plurality of scale feature maps, fusing the plurality of scale feature maps by a multi-scale feature fusion part, and then calculating a category loss function, wherein the category loss function calculation formula is as follows:
where β is the number of background pixels divided by the total number of pixels.
In the description herein, references to the description of "one embodiment," "an example," "a specific example" or the like are intended to mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
The preferred embodiments of the invention disclosed above are intended to be illustrative only. The preferred embodiments are not intended to be exhaustive or to limit the invention to the precise embodiments disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best utilize the invention. The invention is limited only by the claims and their full scope and equivalents.
Claims (1)
1. A steel coil end face edge detection method based on deep learning is characterized by comprising the following steps:
s1, improving the structure of the deep convolutional neural network, and respectively improving the multi-scale feature extraction part and the multi-scale feature fusion part;
s101, a multi-scale feature extraction part consists of five feature extraction modules with different scales, each module comprises two to three convolution layers with convolution kernels of 3 x 3 and a maximum pooling layer with kernels of 2 x 2 and step length of 2, the first scale feature extraction module comprises two convolution layers, the number of convolution filters is 64, the second scale feature extraction module comprises two convolution layers, the number of convolution filters is 128, the subsequent three scale feature extraction modules all comprise three convolution layers, the number of convolution filters is sequentially multiplied by 2 along with scale change, and the scale of a feature map is halved after each pooling layer is passed, so that 5 feature extraction modules output feature maps with 5 different scales;
s102, the multi-scale feature fusion part mainly comprises a convolution layer and a deconvolution layer, for each scale of feature graph, firstly, the feature graph passes through a 1 x 1 convolution layer, then passes through a deconvolution layer to obtain 5 feature graphs with the same size, the 5 feature graphs are spliced together, then pass through a 1 x 1 convolution layer, the function of the convolution layer is to automatically learn the weights of the feature graphs with different scales, and finally the 5 feature graphs are subjected to weighted summation to obtain the output of the network;
s2, collecting the image, sending the collected image into a deep convolution neural network for edge detection, and then outputting an edge detection result;
the edge detection steps are as follows:
inputting a training sample, carrying out multi-scale feature extraction on the training sample to obtain a plurality of scale feature maps, fusing the plurality of scale feature maps by a multi-scale feature fusion part, and then calculating a category loss function, wherein the category loss function calculation formula is as follows:
where β is the number of background pixels divided by the total number of pixels.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910844889.4A CN110648316B (en) | 2019-09-07 | 2019-09-07 | Steel coil end face edge detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910844889.4A CN110648316B (en) | 2019-09-07 | 2019-09-07 | Steel coil end face edge detection method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110648316A CN110648316A (en) | 2020-01-03 |
| CN110648316B true CN110648316B (en) | 2021-02-26 |
Family
ID=69010187
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910844889.4A Active CN110648316B (en) | 2019-09-07 | 2019-09-07 | Steel coil end face edge detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110648316B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112036437B (en) * | 2020-07-28 | 2024-06-07 | 农业农村部南京农业机械化研究所 | Rice seedling detection model based on improved YOLOV network and method thereof |
| CN111915634A (en) * | 2020-08-05 | 2020-11-10 | 武汉伊特森科技有限公司 | Target object edge detection method and system based on fusion strategy |
| CN112581486A (en) * | 2020-11-27 | 2021-03-30 | 深圳点猫科技有限公司 | Edge detection method, device and equipment based on bidirectional cascade network |
| CN112580661B (en) * | 2020-12-11 | 2024-03-08 | 江南大学 | Multi-scale edge detection method under deep supervision |
| CN112950545A (en) * | 2021-02-03 | 2021-06-11 | 南京耘瞳科技有限公司 | Bar steel material bending detection method |
| CN114061524B (en) * | 2021-10-15 | 2024-07-12 | 首钢集团有限公司 | Steel coil contour measuring method and device |
| CN114581806B (en) * | 2022-03-18 | 2024-03-19 | 重庆科技学院 | Industrial part empty rate calculation method based on trunk edge feature extraction |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109522958A (en) * | 2018-11-16 | 2019-03-26 | 中山大学 | Based on the depth convolutional neural networks object detection method merged across scale feature |
Family Cites Families (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016119076A1 (en) * | 2015-01-27 | 2016-08-04 | Xiaoou Tang | A method and a system for face recognition |
| CN107610140A (en) * | 2017-08-07 | 2018-01-19 | 中国科学院自动化研究所 | Near edge detection method, device based on depth integration corrective networks |
| CN107527069A (en) * | 2017-08-22 | 2017-12-29 | 京东方科技集团股份有限公司 | Image processing method, device, electronic equipment and computer-readable medium |
| CN108230329B (en) * | 2017-12-18 | 2021-09-21 | 孙颖 | Semantic segmentation method based on multi-scale convolution neural network |
| CN108830855B (en) * | 2018-04-02 | 2022-03-25 | 华南理工大学 | Full convolution network semantic segmentation method based on multi-scale low-level feature fusion |
| CN108805889B (en) * | 2018-05-07 | 2021-01-08 | 中国科学院自动化研究所 | Edge-guided segmentation method, system and equipment for refined salient objects |
| CN108710919A (en) * | 2018-05-25 | 2018-10-26 | 东南大学 | A kind of crack automation delineation method based on multi-scale feature fusion deep learning |
| CN109410144B (en) * | 2018-10-31 | 2020-11-27 | 聚时科技(上海)有限公司 | End-to-end image defogging processing method based on deep learning |
| CN109711406A (en) * | 2018-12-25 | 2019-05-03 | 中南大学 | A kind of multidirectional image Method for text detection based on multiple dimensioned rotation anchor mechanism |
| CN110084294A (en) * | 2019-04-18 | 2019-08-02 | 北京师范大学 | A kind of Remote Image Classification based on multiple dimensioned depth characteristic |
| CN110084249A (en) * | 2019-04-24 | 2019-08-02 | 哈尔滨工业大学 | The image significance detection method paid attention to based on pyramid feature |
-
2019
- 2019-09-07 CN CN201910844889.4A patent/CN110648316B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109522958A (en) * | 2018-11-16 | 2019-03-26 | 中山大学 | Based on the depth convolutional neural networks object detection method merged across scale feature |
Non-Patent Citations (1)
| Title |
|---|
| Multi-scale object detection in remote sensing imagery with convolutional neural networks;Zhipeng Deng.et;《ISPRS Journal of Photogrammetry and Remote Sensing》;20180502;第145卷;第3-22页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110648316A (en) | 2020-01-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110648316B (en) | Steel coil end face edge detection method based on deep learning | |
| CN111257341B (en) | Underwater building crack detection method based on multi-scale features and stacked full convolution network | |
| CN111429403B (en) | Automobile gear finished product defect detection method based on machine vision | |
| CN108416292B (en) | Unmanned aerial vehicle aerial image road extraction method based on deep learning | |
| CN109034184B (en) | Grading ring detection and identification method based on deep learning | |
| CN111160249A (en) | Multi-class target detection method of optical remote sensing image based on cross-scale feature fusion | |
| CN107564009B (en) | Outdoor scene multi-target segmentation method based on deep convolutional neural network | |
| CN112465759A (en) | Convolutional neural network-based aeroengine blade defect detection method | |
| CN111325762B (en) | Contour detection method based on dense connection decoding network | |
| CN112132196B (en) | Cigarette case defect identification method combining deep learning and image processing | |
| CN113781510B (en) | Edge detection method and device and electronic equipment | |
| CN111639668A (en) | Crowd density detection method based on deep learning | |
| CN111161224A (en) | Casting internal defect grading evaluation system and method based on deep learning | |
| CN111368825A (en) | Pointer positioning method based on semantic segmentation | |
| CN112258537B (en) | Method for monitoring dark vision image edge detection based on convolutional neural network | |
| CN112580661B (en) | Multi-scale edge detection method under deep supervision | |
| CN113887472B (en) | Remote sensing image cloud detection method based on cascade color and texture feature attention | |
| CN109872326B (en) | Contour detection method based on deep reinforced network jump connection | |
| CN113409355A (en) | Moving target identification system and method based on FPGA | |
| CN115049640B (en) | Road crack detection method based on deep learning | |
| Leavline et al. | On teaching digital image processing with MATLAB | |
| CN109859222A (en) | Edge extracting method and system based on cascade neural network | |
| CN111368856A (en) | Spine extraction method and device of book checking system based on vision | |
| CN108932471A (en) | A kind of vehicle checking method | |
| CN114359149A (en) | Dam bank dangerous case video detection method and system based on real-time image edge enhancement |
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 |