CN109840899A - A kind of roughness grade number recognition methods based on depth convolutional neural networks - Google Patents
A kind of roughness grade number recognition methods based on depth convolutional neural networks Download PDFInfo
- Publication number
- CN109840899A CN109840899A CN201811559998.3A CN201811559998A CN109840899A CN 109840899 A CN109840899 A CN 109840899A CN 201811559998 A CN201811559998 A CN 201811559998A CN 109840899 A CN109840899 A CN 109840899A
- Authority
- CN
- China
- Prior art keywords
- neural networks
- convolutional neural
- roughness
- depth convolutional
- grade number
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention proposes a kind of roughness grade number recognition methods based on depth convolutional neural networks, and the roughness grade number image data base of the roughness standards block including establishing different processing methods, trains depth convolutional neural networks to obtain depth convolutional neural networks model and by roughness grade number image input depth convolutional neural networks model to obtain roughness grade number at building depth convolutional neural networks.The present invention utilizes depth convolutional neural networks, can recognize the roughness of a variety of processing methods, solves the problems, such as that Conventional visual method can only differentiate the roughness grade number of single processing method.
Description
Technical field
The invention belongs to roughness grade number identification field more particularly to a kind of roughness based on depth convolutional neural networks
Grade recognition methods.
Background technique
Surface roughness is to evaluate an important indicator of workpiece surface quality, influences service life and the service performance of workpiece.
With the raising of Automation in Mechanical Working degree, numerous parts are changed to examine by inspecting by random samples, propose to surface roughness on-line measurement
Increasingly higher demands, therefore in production process real-time detection surface roughness is increasingly taken seriously.Traditional tracer method is wanted
The control for guaranteeing measuring force size, should guarantee that gauge head contacts always with surface, and cannot therefore scratch workpiece surface and abrasion
Gauge head.So traditional needle contact method has limitation.
Radiation of visible light can occur to reflect and be reflected in body surface, and the intensity for reflecting and reflecting depends on object table
The roughness in face, therefore the textural characteristics of the piece surface image taken with CCD camera and roughness have close ties.It calculates
Machine vision technique is reached with the classifier of artificial intelligence field to surface by extracting the textural characteristics of piece surface image
The purpose of roughness grade number classification.But these methods all can only differentiate the roughness grade number of single processing method and cannot achieve pair
The roughness grade number of a variety of processing methods identifies.
Summary of the invention
The purpose of the present invention is to provide a kind of roughness grade number recognition methods based on depth convolutional neural networks, the party
Method can identify the roughness grade number of a variety of processing methods simultaneously.To achieve the above object, the technical scheme adopted by the invention is as follows:
A kind of roughness grade number recognition methods based on depth convolutional neural networks, comprising the following steps:
S1: the roughness grade number image data base of the roughness standards block of different processing methods is established;
S2: building depth convolutional neural networks;
S3: training depth convolutional neural networks are to obtain depth convolutional neural networks model;
S4: by roughness grade number image input depth convolutional neural networks model to obtain roughness grade number.
Preferably, step S1 specifically:
S11: roughness standards block is obtained to the roughness grade number image of identical amplification factor under the microscope;
S12: roughness grade number label is prepared according to roughness grade number image;
S13: roughness grade number image is pre-processed for training depth convolutional neural networks.
Preferably, processing method includes plain milling, vertical milling, turning, plain grinding and grinding in step S1.
Preferably, the processing method respectively includes the roughness standards block of multiple roughness grade numbers.
Preferably, in step S11, the different zones of roughness standards block are sampled.
Preferably, in step S13, it is 227*227 that the roughness grade number image, which carries out pretreated Pixel Dimensions,.
Preferably, in step S2, the depth convolutional neural networks include 5 layers of convolutional layer that successively signal connects, 3 layers of pond
Change layer and 3 layers of full articulamentum;The input layer of the depth convolutional neural networks is to receive rugosity level images;Described 3 layers connect entirely
The last layer for connecing layer is output layer.
Preferably, in step S3, training method includes that the propagated forward of parameter is trained and error back propagation is trained;It is described
Propagated forward training refers to that roughness grade number image, which is input to the depth convolutional neural networks, obtains practical roughness of network etc.
Grade recognition result;
The error back propagation training specifically: first calculate the practical roughness grade number recognition result of network and it is expected coarse
The error between grade recognition result is spent, then backpropagation calculates each layer of error, and successively updates each layer parameter, so follows
Ring, until completing whole iteration.
Preferably, the propagated forward is trained and error back propagation training is all made of stochastic gradient descent method to the depth
Spend convolutional neural networks iteration.
Compared with prior art, advantages of the present invention are as follows: utilize depth convolutional neural networks, can recognize a variety of processing methods
Roughness, solve the problems, such as that Conventional visual method can only differentiate the roughness grade number of single processing method.The present invention simultaneously
It eliminates and feature extraction and screening is carried out to roughness grade number image, process is simple, convenient for operation.
Detailed description of the invention
Fig. 1 is training and the depth convolutional neural networks model for the depth convolutional neural networks that one embodiment of the invention provides
Identify the flow chart of roughness grade number;
Fig. 2 is roughness grade number image sampling schematic diagram;
Fig. 3 is the training general flow chart of depth convolutional neural networks;
Fig. 4 is the roughness grade number identification process figure of depth convolutional neural networks model;
Fig. 5 is the network structure and parameter of depth convolutional neural networks;
Fig. 6~Figure 13 is the roughness grade number image under different processing methods;
Figure 14 is the recognition accuracy curve of depth convolutional neural networks model;
Figure 15 is the error curve of the training of depth convolutional neural networks and the identification of depth convolutional neural networks model.
1- roughness standards block, the first pickup area of 11-, the second pickup area of 12-.
Specific embodiment
Below in conjunction with schematic diagram to the roughness grade number recognition methods of the invention based on depth convolutional neural networks, into
Row more detailed description, which show the preferred embodiment of the present invention, it should be appreciated that those skilled in the art can modify
The present invention of this description, and still realize advantageous effects of the invention.Therefore, following description should be understood as this field
Technical staff's is widely known, and is not intended as limitation of the present invention.
This implementation proposes a kind of roughness grade number recognition methods based on depth convolutional neural networks, includes the following steps
S1~S4, specific as follows:
S1: the roughness grade number image data base of the roughness standards block 1 of different processing methods is established;Processing method includes
Plain milling, vertical milling, turning, plain grinding and grinding, each processing method respectively include the roughness standards block 1 of multiple roughness grade numbers.
S11: roughness standards block 1 is obtained to the roughness grade number image of identical amplification factor under Stereo microscope;Slightly
Rugosity level images specifically: plain milling (Ra6.3, Ra3.2, Ra1.6, Ra0.8, Ra0.4), vertical milling (Ra6.3, Ra3.2,
Ra1.6, Ra0.8, Ra0.4), turning (Ra6.3, Ra3.2, Ra1.6, Ra0.8, Ra0.4), plain grinding (Ra0.05, Ra0.1,
Ra0.2, Ra0.4, Ra0.8, Ra1.6), it grinds (Ra0.05, Ra0.1).Since roughness grade number image data base needs are more
Data carry out multiple repairing weld to the different zones of roughness standards block 1, and roughness standards block 1 includes 11 He of the first pickup area
Second pickup area 12, it is specific as shown in Figure 2.As shown in Fig. 6~13: Fig. 6 is from left to right represented sequentially as turning Ra6.3, puts down
Mill the roughness grade number image of Ra6.3, vertical milling Ra6.3;Fig. 7 is from left to right represented sequentially as turning Ra3.2, plain milling Ra3.2, stands
Mill Ra3.2 roughness grade number image;Fig. 8 is from left to right represented sequentially as turning Ra1.6, plain milling Ra1.6, vertical milling Ra1.6, plain grinding
Ra1.6, cylindrical grinding Ra1.6 roughness grade number image;Fig. 9 is from left to right represented sequentially as turning Ra0.8, plain milling Ra0.8, vertical milling
Ra0.8, plain grinding Ra0.8, cylindrical grinding Ra0.8 roughness grade number image;Figure 10 is from left to right represented sequentially as turning Ra0.4, puts down
Mill Ra0.4, vertical milling Ra0.4, plain grinding Ra0.4, cylindrical grinding Ra0.4 roughness grade number image;Figure 11 is from left to right represented sequentially as
Plain grinding Ra0.2, cylindrical grinding Ra0.2 roughness grade number image;Figure 12 is from left to right represented sequentially as plain grinding Ra0.1, grinding Ra0.1
Roughness grade number image;Figure 13 is from left to right represented sequentially as plain grinding Ra0.05, grinding Ra0.05 roughness grade number image.
S12: roughness grade number label is prepared according to roughness grade number image;
S13: roughness grade number image is pre-processed for training depth convolutional neural networks, roughness grade number figure
It is 227*227 as carrying out pretreated Pixel Dimensions.
S2: building depth convolutional neural networks;Depth convolutional neural networks include 5 layers of convolutional layer, 3 of successively signal connection
Layer pond layer and 3 layers of full articulamentum;The input layer of depth convolutional neural networks is to receive rugosity level images;3 layers of full articulamentum
The last layer be output layer.The structure and parameter of the depth convolutional neural networks are as shown in Figure 5.Depth in network training process
The calculating for spending the convolution sum sampling process of convolutional neural networks is as follows:
1) convolution layer parameter calculates
Roughness grade number seal is X'i(H × K), H=K=227 determine convolution kernel Xs(w × h), training one are sparse certainly
Dynamic encoder, learns k feature:
fs=σ (W(1)Xs+b(1)) (11)
In formula: σ is Sigmoid type function, W(1)And b(1)It is aobvious weight and biasing of the layer unit to Hidden unit, f respectivelys
It is a convolution Feature Mapping, size calculates are as follows:
S(fs)=k × [((H+2 × pad-w)/stride)+1] × [((K+2 × pad-h)/stride)+1] (12)
In formula, k is convolution kernel number;Pad is border extended parameter, default value 0;Stride is convolution kernel step-length, is write from memory
Think 1.
2) pond layer parameter calculates
To extract high-level characteristic, pondization operation need to be carried out.Chi Huahe is having a size of kp(m × n), divide convolutional layer be it is several not
The region m × n of intersection carries out pondization operation.Pond method includes maximum value pond, average value pond and random value pond, this paper
Maximum value pond is selected, if fs(r × c) is a convolution Feature Mapping, maximum value pond formula are as follows:
ps=maxm×n(fs)
The Feature Mapping size calculation formula of Chi Huahou are as follows:
S(ps)=k × [((r+2 × pad-m)/stride)+1] × [((c+2 × pad-n)/stride)+1]
In formula, k is Chi Huahe number;Pad is border extended number, default value 0;Stride is Chi Huahe step-length, is write from memory
Recognizing value is 1.
S3: training depth convolutional neural networks are to obtain depth convolutional neural networks model;Training method includes parameter
Propagated forward is trained and error back propagation is trained;Propagated forward training, which refers to, is input to depth convolution mind for roughness grade number image
The practical roughness grade number recognition result of network is obtained through network;Error back propagation training specifically: it is actually coarse to calculate network
It spends grade recognition result and it is expected that the error between roughness grade number recognition result, then backpropagation calculate each layer of error,
And each layer parameter is successively updated, it so recycles, until completing whole iteration.Error back propagation training uses stochastic gradient descent
For method to depth convolutional neural networks iteration, every iteration once calculates a gradient value, finds out in depth convolutional neural networks every layer
The weight and biasing optimal solution of network, training obtains depth convolutional neural networks model to successive ignition later.As shown in figure 3, being
The training general flow chart of depth convolutional neural networks.
S4: by roughness grade number image input depth convolutional neural networks model to obtain roughness grade number.Wherein Fig. 1 is
The flow chart of step S3~S4, Fig. 4 are the flow chart of step S4.Figure 14 is that the identification of roughness grade number in training process is accurate
Rate, wherein abscissa is the number of iterations, and ordinate is precision.By experimental verification, roughness grade number accuracy of judgement degree of the present invention
Reach 99.6%;Figure 15 is the error curve of roughness grade number identification in trained and test process, wherein abscissa is iteration time
Number, ordinate is error.C2 indicates training process error with the number of iterations change curve, and C1 indicates test process error with iteration
Number change curve.
The above is only a preferred embodiment of the present invention, does not play the role of any restrictions to the present invention.Belonging to any
Those skilled in the art, in the range of not departing from technical solution of the present invention, to the invention discloses technical solution and
Technology contents make the variation such as any type of equivalent replacement or modification, belong to the content without departing from technical solution of the present invention, still
Within belonging to the scope of protection of the present invention.
Claims (9)
1. a kind of roughness grade number recognition methods based on depth convolutional neural networks, which comprises the following steps:
S1: the roughness grade number image data base of the roughness standards block of different processing methods is established;
S2: building depth convolutional neural networks;
S3: training depth convolutional neural networks are to obtain depth convolutional neural networks model;
S4: by roughness grade number image input depth convolutional neural networks model to obtain roughness grade number.
2. the roughness grade number recognition methods according to claim 1 based on depth convolutional neural networks, which is characterized in that
Step S1 specifically:
S11: roughness standards block is obtained to the roughness grade number image of identical amplification factor under the microscope;
S12: roughness grade number label is prepared according to roughness grade number image;
S13: roughness grade number image is pre-processed for training depth convolutional neural networks.
3. the roughness grade number recognition methods according to claim 2 based on depth convolutional neural networks, which is characterized in that
Processing method includes plain milling, vertical milling, turning, plain grinding and grinding in step S1.
4. the roughness grade number recognition methods according to claim 3 based on depth convolutional neural networks, which is characterized in that
The processing method respectively includes the roughness standards block of multiple roughness grade numbers.
5. the roughness grade number recognition methods according to claim 2 based on depth convolutional neural networks, which is characterized in that
In step S11, the different zones of roughness standards block are sampled.
6. the roughness grade number recognition methods according to claim 2 based on depth convolutional neural networks, which is characterized in that
In step S13, it is 227*227 that the roughness grade number image, which carries out pretreated Pixel Dimensions,.
7. the roughness grade number recognition methods according to claim 1 based on depth convolutional neural networks, which is characterized in that
In step S2, the depth convolutional neural networks include that the 5 layers of convolutional layer, 3 layers of pond layer of successively signal connection and 3 layers connect entirely
Layer;The input layer of the depth convolutional neural networks is to receive rugosity level images;The last layer of described 3 layers full articulamentum is
Output layer.
8. the roughness grade number recognition methods according to claim 7 based on depth convolutional neural networks, which is characterized in that
In step S3, training method includes that the propagated forward of parameter is trained and error back propagation is trained;The propagated forward training refers to
Roughness grade number image is input to the depth convolutional neural networks and obtains the practical roughness grade number recognition result of network;
The error back propagation training specifically: first calculate the practical roughness grade number recognition result of network and desired roughness etc.
Error between grade recognition result, then backpropagation calculate each layer of error, and successively update each layer parameter, so recycle,
Until completing whole iteration.
9. the roughness grade number recognition methods according to claim 8 based on depth convolutional neural networks, which is characterized in that
The propagated forward is trained and error back propagation training is all made of stochastic gradient descent method to the depth convolutional neural networks
Iteration.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811559998.3A CN109840899A (en) | 2018-12-20 | 2018-12-20 | A kind of roughness grade number recognition methods based on depth convolutional neural networks |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811559998.3A CN109840899A (en) | 2018-12-20 | 2018-12-20 | A kind of roughness grade number recognition methods based on depth convolutional neural networks |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109840899A true CN109840899A (en) | 2019-06-04 |
Family
ID=66883342
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811559998.3A Pending CN109840899A (en) | 2018-12-20 | 2018-12-20 | A kind of roughness grade number recognition methods based on depth convolutional neural networks |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109840899A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110322077A (en) * | 2019-07-10 | 2019-10-11 | 燕山大学 | Cement raw material Vertical Mill raw material fineness index prediction technique based on convolutional neural networks |
CN110470608A (en) * | 2019-08-15 | 2019-11-19 | 杭州电子科技大学 | A kind of method and device using polarization imaging measurement object smoothness |
CN111709924A (en) * | 2020-06-11 | 2020-09-25 | 东北大学 | Intelligent 3D rock structural surface roughness extraction system and method |
CN113034482A (en) * | 2021-04-07 | 2021-06-25 | 山东大学 | Surface roughness detection method based on machine vision and machine learning |
CN114357851A (en) * | 2021-10-23 | 2022-04-15 | 西北工业大学 | Numerical control milling surface roughness prediction method based on DAE-RNN |
CN114511528A (en) * | 2022-01-25 | 2022-05-17 | 上海理工大学 | Cloud edge cooperation-based workpiece surface roughness online detection method and system |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108734142A (en) * | 2018-05-28 | 2018-11-02 | 西南交通大学 | A kind of core in-pile component surface roughness appraisal procedure based on convolutional neural networks |
-
2018
- 2018-12-20 CN CN201811559998.3A patent/CN109840899A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108734142A (en) * | 2018-05-28 | 2018-11-02 | 西南交通大学 | A kind of core in-pile component surface roughness appraisal procedure based on convolutional neural networks |
Non-Patent Citations (2)
Title |
---|
CHA Y.J 等: "Deep Learning Based Crack Damage Detection Using Convolutional Neural Networks", 《COMPUTER-AIDED CIVIL&INFRASTRUCTURE ENGINEERING》 * |
WEIFANG SUN 等: "Noncontact Surface Roughness Estimation Using 2D Complex Wavelet Enhanced ResNet for Intelligent Evaluation of Milled Metal Surface Quality", 《MDPI》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110322077A (en) * | 2019-07-10 | 2019-10-11 | 燕山大学 | Cement raw material Vertical Mill raw material fineness index prediction technique based on convolutional neural networks |
CN110322077B (en) * | 2019-07-10 | 2022-08-02 | 燕山大学 | Cement raw material vertical mill raw material fineness index prediction method based on convolutional neural network |
CN110470608A (en) * | 2019-08-15 | 2019-11-19 | 杭州电子科技大学 | A kind of method and device using polarization imaging measurement object smoothness |
CN111709924A (en) * | 2020-06-11 | 2020-09-25 | 东北大学 | Intelligent 3D rock structural surface roughness extraction system and method |
CN113034482A (en) * | 2021-04-07 | 2021-06-25 | 山东大学 | Surface roughness detection method based on machine vision and machine learning |
CN114357851A (en) * | 2021-10-23 | 2022-04-15 | 西北工业大学 | Numerical control milling surface roughness prediction method based on DAE-RNN |
CN114511528A (en) * | 2022-01-25 | 2022-05-17 | 上海理工大学 | Cloud edge cooperation-based workpiece surface roughness online detection method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109840899A (en) | A kind of roughness grade number recognition methods based on depth convolutional neural networks | |
CN108520274B (en) | High-reflectivity surface defect detection method based on image processing and neural network classification | |
CN110570393A (en) | mobile phone glass cover plate window area defect detection method based on machine vision | |
CN105069790B (en) | A kind of gear open defect fast image detection method | |
CN109900706A (en) | A kind of weld seam and weld defect detection method based on deep learning | |
CN106875373B (en) | Mobile phone screen MURA defect detection method based on convolutional neural network pruning algorithm | |
CN104981105B (en) | A kind of quickly accurate detection and method for correcting error for obtaining element central and deflection angle | |
CN113362326A (en) | Method and device for detecting welding spot defects of battery | |
CN111189837B (en) | Cigarette appearance online detection method and device | |
CN102496161B (en) | Method for extracting contour of image of printed circuit board (PCB) | |
CN109919934A (en) | A kind of liquid crystal display panel defect inspection method based on the study of multi-source domain depth migration | |
CN112037219A (en) | Metal surface defect detection method based on two-stage convolution neural network | |
CN108985337A (en) | A kind of product surface scratch detection method based on picture depth study | |
CN112233067A (en) | Hot rolled steel coil end face quality detection method and system | |
CN109540925B (en) | Complex ceramic tile surface defect detection method based on difference method and local variance measurement operator | |
CN109444172A (en) | A kind of automatic optical detection device and its method suitable for artificial intelligence detection PCB | |
CN109685793A (en) | A kind of pipe shaft defect inspection method and system based on three dimensional point cloud | |
CN113222901A (en) | Method for detecting surface defects of steel ball based on single stage | |
CN111754463A (en) | Ballastless track CA mortar layer defect detection method based on convolutional neural network | |
CN116612106A (en) | Method for detecting surface defects of optical element based on YOLOX algorithm | |
CN111260609A (en) | Cigarette appearance defect detection method based on deep learning | |
CN115937518A (en) | Pavement disease identification method and system based on multi-source image fusion | |
CN114881998A (en) | Workpiece surface defect detection method and system based on deep learning | |
CN109886936B (en) | Low-contrast defect detection method and device | |
CN113538342B (en) | Convolutional neural network-based aluminum aerosol can coating quality detection method |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190604 |