CN109215015A - A kind of online visible detection method of silk cocoon based on convolutional neural networks - Google Patents
A kind of online visible detection method of silk cocoon based on convolutional neural networks Download PDFInfo
- Publication number
- CN109215015A CN109215015A CN201810820347.9A CN201810820347A CN109215015A CN 109215015 A CN109215015 A CN 109215015A CN 201810820347 A CN201810820347 A CN 201810820347A CN 109215015 A CN109215015 A CN 109215015A
- Authority
- CN
- China
- Prior art keywords
- silk cocoon
- image
- layer
- convolutional neural
- neural networks
- 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
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
- G06T7/001—Industrial image inspection using an image reference approach
-
- 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]
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of online visible detection methods of the silk cocoon based on convolutional neural networks, carry out region selection to the silk cocoon image of acquisition, and category constructs silk cocoon data set；Simultaneously training convolutional neural networks are designed, the silk cocoon disaggregated model that accuracy of identification is high and parameter amount is small is formed；Captured in real-time picture in on-line checking, chooses silk cocoon region automatically, is input in disaggregated model and is predicted, obtains classification results；Network model replaces the full articulamentum in traditional convolution neural network structure using global average pond layer, it is few with parameter, used time short feature is tested, can be realized the quick detection to silk cocoon, the pressure of artificial screening silk cocoon is greatly reduced and there is stronger adaptive ability.
Description
Technical field
The silk cocoon online test method based on convolutional neural networks that the present invention relates to a kind of, belongs to field of machine vision.
Background technique
Silk is the rarity of China, and the economic benefit of annual silk industry is very high, and the yield of related silk product is huge
Greatly, annual demand is all continuously increased.Raw material of the silk cocoon as silk, the superiority and inferiority of quality also generate very silk product quality
It is big to influence, therefore it is most important to production high quality silk product to filter out good silk cocoon.
The mode that hand picking is generallyd use in silk cocoon industry at present is screened, this mode not only low efficiency, but also
Judgment criteria relies primarily on the experience of master worker, and for the result of screening vulnerable to emotion influence, this results in finally obtained silk cocoon
Possible standard has difference, influences the subsequent operations such as filature.
The efficiency to work in order to further increase, the demand that silk cocoon automates assisting sifting are increasing.Automation auxiliary
Screening mainly includes transmission device, vision detection system and sorting equipment three parts.It will be closed by Machine Vision Detection algorithm
Lattice silk cocoon and defect silk cocoon carry out the key technology that classification is silk cocoon automatically screening.
Currently, there is related technical personnel to carry out some researchs to silk cocoon detection using the method for machine vision, wherein Song
Sub- outstanding person et al. proposes to calculate the features such as silk cocoon area, related coefficient and rgb value according to the method for mathematical morphology to judge silk cocoon
Type, but every kind of silk cocoon needs to classify using different parameters, and every kind of parameter needs to adjust, and process is complex, and by
The many factors of light, image processing method etc. influence.The peak Zhejiang University Jin Hang uses image binaryzation, contrast expansion, figure
As a variety of digital image processing techniques such as reverse phases, silk cocoon image is handled and extracts silk cocoon surface region, by area
Domain carries out characteristic parameter extraction, is based ultimately upon multiple parameters and identifies to silk cocoon, but its accuracy also still has from field application
Gap, the certain features for being primarily due to cocoon are close with peripheral region color, although using the method for enhancing contrast, but nothing
Method extracts characteristic area completely.
The patent of Publication No. CN106906520A discloses " method and silk cocoon the screening system of the inferior silk cocoon of electronic recognition
System ", this method first identifies the profile of silk cocoon by the method for machine vision, carries out silkworm again after silk cocoon outline identification
The identification of cocoon pixel and color, if profile is more than or less than setting value or silk cocoon identifying system recognizes the pixel or face of setting
Color belongs to identification types, then silk cocoon is identified camera and is labeled as inferior silk cocoon.It, should under conditions of setting value is in reasonable interval
Method can reduce the pressure of manual identified to a certain extent.However, setting value described in this method is similarly subjected to background, lamp
The influence of the multiple factors such as light, once environment changes, setting value just needs to adjust repeatedly, and otherwise the selection result of silk cocoon will produce
It is raw to influence.
In recent years, with the fast development of deep learning, the automatic of feature is carried out to image using convolutional neural networks
It extracts and the method for classification is increasingly paid close attention to by people.This method makes image avoid traditional knowledge directly as the input of network
Complicated characteristic extraction procedure in other algorithm, and have good generalization ability, to the geometric transformation occurred in identification process,
Deformation, uneven illumination is even good adaptive ability.
Summary of the invention
The silk cocoon online test method based on convolutional neural networks that it is an object of that present invention to provide a kind of is realized to inhomogeneity
Other silk cocoon accurately identifies.
To reach this purpose, technical scheme is as follows: carrying out region selection to the silk cocoon image of acquisition, and presses class
It Gou Jian not silk cocoon data set；Simultaneously training convolutional neural networks are designed, the silk cocoon classification mould that accuracy of identification is high and parameter amount is small is formed
Type；Captured in real-time picture in on-line checking, chooses silk cocoon region automatically, is input in disaggregated model and is predicted, is classified
As a result, specifically includes the following steps:
S1 establishes silk cocoon data set
S1.1 obtains different classes of silk cocoon image, and different classes of silk cocoon image is shot under dark-background, constructs silk cocoon
Image original image library.
S1.2 carries out gaussian filtering process in the original image of acquisition, executes binaryzation, enhancing comparison to filtered image
Degree.
S1.3 seeks the barycentric coodinates of silk cocoon region in binary map.
S1.4 is extended in original image centered on center of gravity and intercepts out the region of W*H to remove extra background information, made this
Region can completely include all kinds of silk cocoons, wherein W, and H is fixed numbers, and specific value is carried out according to silk cocoon in image proportion
Adjustment.Silk cocoon image data set is established according to step S1.1-S1.4.
S2 is designed and is trained the silk cocoon disaggregated model based on convolutional neural networks:
Training silk cocoon disaggregated model carries out established silk cocoon image data set using the method for convolutional neural networks
Training, trained process are broadly divided into three steps:
The adjustment of S2.1 picture size and normalized.The size for adjusting image in silk cocoon data set first is set as solid
Determine size, pixel normalized secondly is carried out to image.
S2.2 design is directed to the convolutional neural networks structure of silk cocoon.The convolutional neural networks built use eight layers of structure, preceding
Four layers are alternately arranged for convolutional layer with pond layer, and layer 5 and layer 6 are convolutional layer, and layer 7 is global average pond layer, the
Eight layers are classification layer, and every layer of weight realizes initialization using the random number of Gaussian Profile.As shown in Fig. 2, being silkworm of the invention
Cocoon convolutional neural networks structure.Conv1 layers, Conv2 layers, Conv3 layers, Conv4 layers have 64,128,192,192 respectively
Feature map, the size of all convolution kernels are 3x3, and Pool1 layers, Pool2 layers are maximum pond layer, GlobalAVE
Pool is global average pond layer, and the last layer is Softmax classification layer.
S2.3) convolutional neural networks of design are trained.In the training process, small batch (mini is used every time
A certain number of silk cocoon training samples are sent into network by mode batch), are finally exported using Softmax classification layer each
The prediction result of sample, by multinomial logic loss function (Multinomial Logistic Loss) measure prediction result with
Gap between legitimate reading uses the gradient descent method of momentum-type so that loss function is reached minimum for optimization method, Jin Ershi
Now to the training of convolutional neural networks, the convolutional neural networks disaggregated model for silk cocoon classification is obtained.
S3 on-line checking process:
S3.1 silk cocoon image) is obtained online, is adjusted the parameters such as industrial camera lens aperture, focal length, is pacified on the outside of camera
Annular light source is filled, shoots silk cocoon image under dark-background.
S3.2) for the silk cocoon image shot under the dark-background of acquisition, using the method for template matching, Primary Location goes out
Under dark-background in captured silk cocoon image silk cocoon position.
S3.3) center of silk cocoon in silk cocoon image captured under dark-background is extended, is expanded to specified big
It is small for screenshot after W × H, so that only one silk cocoon on every screenshot is located at the center of image.
S3.4 the size for) adjusting every screenshot is fixed size.
S3.5) subtract the method for normalizing of mean value to having adjusted the screenshot after size and executed, the image after being normalized.
S3.6) image after normalization is sent into trained CNN model, by the propagated forward of model, is obtained pre-
It surveys as a result, obtaining the classification information of silk cocoon.
The beneficial effects of the present invention are: the present invention provides a kind of silk cocoon on-line checking side based on convolutional neural networks
Method.This method mainly includes that silk cocoon data set is established, three major parts of off-line training and on-line checking.Firstly the need of establishing silkworm
Then cocoon data set constructs simultaneously training convolutional neural networks silk cocoon disaggregated model as the data source of model training.Online inspection
Captured in real-time picture in survey, chooses silk cocoon region automatically, is input in disaggregated model and is predicted, obtains classification results.Home Network
Network model replaces the full articulamentum in traditional convolution neural network structure using global average pond layer, has parameter few, test
Used time short feature, can be realized the quick detection to silk cocoon, the pressure of artificial screening silk cocoon is greatly reduced, while can expire
High real-time in sufficient engineering, the requirement of high-accuracy, and there is stronger adaptive ability.
Detailed description of the invention
System overall flow figure Fig. 1 of the invention.
Training convolutional neural networks disaggregated model overall structure Fig. 2 of the invention.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention more comprehensible with reference to specific embodiments and reference
Attached drawing, invention is further described in detail.It should be appreciated that specific embodiment described herein is only to explain this hair
It is bright, it is not intended to limit the present invention.
The present invention is a kind of silk cocoon category detection method, and this method mainly includes the foundation of silk cocoon data set, off-line training
With three processes of on-line checking, system overall flow figure such as Fig. 1.This method carries out region selection to collected silk cocoon image,
And category constructs silk cocoon data set；Design and training convolutional neural networks silk cocoon disaggregated model, model overall structure such as Fig. 2,
Form the disaggregated model that accuracy of identification is high and parameter amount is small；Captured in real-time picture in on-line checking chooses silk cocoon region automatically, defeated
Enter and predicted into disaggregated model, obtains classification results.
Further, the specific implementation step of silk cocoon data set is established are as follows:
Step 1: the acquisition of silk cocoon image；
According to silk cocoon recognition detection required precision, the mechanical parameters such as industrial camera lens aperture, focal length are adjusted, obtain silkworm
Cocoon image.
Step 2: picture size pre-processes；
(1) gaussian filtering process is carried out in the original image of acquisition, and binaryzation, enhancing comparison are executed to filtered image
Degree.
(2) barycentric coodinates (x, y) of silk cocoon region in binary map are sought.
(3) it centered on center of gravity, is extended in original image and intercepts out the region of W*H to remove extra background information, make this
Region can completely include all kinds of silk cocoons, wherein W, and H is fixed numbers, and specific value is carried out according to silk cocoon in image proportion
Adjustment.Specific extended method is two coordinates for finding the upper left corner and the lower right corner of target area, and wherein the point in the upper left corner is sat
It is designated as (x-W/2, y-H/2), the point coordinate in the lower right corner is (x+W/2, y+H/2).Screenshot is carried out by the two point coordinates, is obtained
Silk cocoon image after removing background executes step 2 to every obtained original image, establishes silk cocoon image data set.
Further, it designs and silk cocoon disaggregated model of the training based on convolutional neural networks implements step are as follows:
Step 1: carrying out size adjusting and normalized to the sample in silk cocoon data set.
The size that image in silk cocoon data set is adjusted by way of squash, is set as 227*227, to image into
Row pixel normalized, concrete operations are that all silk cocoon images of data concentration are sought with R, the mean value in tri- channels G, B, then
The mean value that tri- channels RGB of every image subtract corresponding channel realizes normalization.
Step 2: design convolutional neural networks structure.
The convolutional neural networks that this method is built use 8 layers of structure, and first four layers are alternately arranged for convolutional layer with pond layer, the
Five layers are convolutional layer with layer 6, and layer 7 is global average pond layer, and the 8th layer is classification layer, and every layer of weight uses Gauss
The random number of distribution realizes initialization.As shown in Fig. 2, being the silk cocoon convolutional neural networks structure of this method.Conv1 layers, Conv2
Layer, Conv3 layers, Conv4 layers be convolutional layer, every layer of convolution kernel size is 3x3, the quantity of convolution kernel is respectively 64,
128,192,192, after each convolution operation to obtained characteristic pattern using ReLU function execute Nonlinear Processing, Pool1 layers,
Pool2 layers are maximum pond layer, and Global AVE Pool is global average pond layer, and the last layer is Softmax classification
Layer.
| Layer name | Layer remarks | Remarks are explained |
| Conv1 | 3x3x64 | Core size x nuclear volume |
| Pool1 | 3x3/2 | Core size/step-length |
| Conv2 | 3x3x128 | Core size x nuclear volume |
| Pool2 | 3x3/2 | Core size/step-length |
| Conv3 | 3x3x192 | Core size x nuclear volume |
| Conv4 | 3x3x192 | Core size x nuclear volume |
| GlobalAVEPool | 29x29 | Core size |
| Softmax | 5 | Output |
Step 3: the convolutional neural networks to design are trained, the convolutional neural networks point for silk cocoon classification are generated
Class model.
(1) in the training process, every time by the way of small batch (mini batch) by the training of a certain number of silk cocoons
Sample is sent into network, the final prediction result that each sample is exported using Softmax classification layer.
(2) by multinomial logic loss function (Multinomial Logistic Loss) measurement prediction result and really
As a result the gap between:
Wherein, E is error loss, wherein pnIndicate that the silk cocoon image n of prediction is the probability of its true tag label, N is
This batch is input to the sum of the silk cocoon image in model.
(3) gradient descent method of momentum-type is used so that loss function is reached minimum for optimization method, wherein momentum parameter is set
It is set to 0.9, and then realizes the training to convolutional neural networks, obtains the convolutional neural networks disaggregated model for silk cocoon classification.
Further, on-line checking process implements step are as follows:
Step 1: Primary Location goes out silk cocoon position in image；
The requirement of real-time online detection accuracy is identified according to silk cocoon, adjusts the parameters such as industrial camera lens aperture, focal length, it is real
When obtain silk cocoon image, intercept the overall profile of qualified cocoon under the present conditions as template, at the beginning of by way of template matching
Step orients each silk cocoon position.
Step 2: extending matching position to specified size；
By obtained silk cocoon preliminary matches information, the center (x of each silk cocoon is calculated1,y1), (x2,y2)……
(xj,yj) and intercept on the basis of center the region of W*H around, j indicates the number of silk cocoon, and W is the width of interception area
Degree, H are the height of interception area.The method of the specific interception rectangular area W*H is to find the upper left corner and the lower right corner of target area
Two coordinates, wherein the point coordinate in the upper left corner is (x-W/2, y-H/2), and the point coordinate in the lower right corner is (x+W/2, y+H/2),
Screenshot is carried out by the upper left corner and two, lower right corner coordinate.
Step 3: image normalization handles and is sent into model；
Truncated picture is adjusted to 227*227 size, then the image subtract the normalized of mean value, will handle
Good image is sent into trained model, by propagated forward, output category information.
Those of ordinary skill in the art will understand that the embodiments described herein, which is to help reader, understands this hair
Bright principle, it should be understood that protection scope of the present invention is not limited to such specific embodiments and embodiments.This field
Those of ordinary skill disclosed the technical disclosures can make according to the present invention and various not depart from the other each of essence of the invention
The specific variations and combinations of kind, these variations and combinations are still within the scope of the present invention.
Claims (4)
1. a kind of online visible detection method of silk cocoon based on convolutional neural networks, it is characterised in that: to the silk cocoon image of acquisition
Region selection is carried out, and category constructs silk cocoon data set；Simultaneously training convolutional neural networks are designed, accuracy of identification height is formed and are joined
The small silk cocoon disaggregated model of quantity；Captured in real-time picture in on-line checking, chooses silk cocoon region automatically, is input in disaggregated model
It is predicted, obtains classification results, specifically includes the following steps:
S1 establishes silk cocoon data set
S1.1 obtains different classes of silk cocoon image, and different classes of silk cocoon image is shot under dark-background, constructs silk cocoon image
Original image library；
S1.2 carries out gaussian filtering process in the original image of acquisition, executes binaryzation to filtered image, enhances contrast；
S1.3 seeks the barycentric coodinates of silk cocoon region in binary map；
S1.4 is extended in original image centered on center of gravity and intercepts out the region of W*H to remove extra background information, made the region
It can completely include all kinds of silk cocoons, wherein W, H are fixed numbers, and specific value is adjusted according to silk cocoon in image proportion
It is whole；Silk cocoon image data set is established according to step S1.1-S1.4；
S2 is designed and is trained the silk cocoon disaggregated model based on convolutional neural networks:
Training silk cocoon disaggregated model is trained established silk cocoon image data set using the method for convolutional neural networks,
Trained process is broadly divided into three steps:
The adjustment of S2.1 picture size and normalized；The size for adjusting image in silk cocoon data set first, is set as fixed ruler
Secondly very little size carries out pixel normalized to image；
S2.2 design is directed to the convolutional neural networks structure of silk cocoon；The convolutional neural networks built use eight layers of structure, first four layers
It being alternately arranged for convolutional layer and pond layer, layer 5 and layer 6 are convolutional layer, and layer 7 is global average pond layer, the 8th layer
For layer of classifying, every layer of weight realizes initialization using the random number of Gaussian Profile；Conv1 layer, Conv2 in network structure
Layer, Conv3 layers, Conv4 layers have 64,128,192,192 feature map respectively, the size of all convolution kernels is 3x3,
Pool1 layers, Pool2 layers are maximum pond layer, and GlobalAVE Pool is global average pond layer, and the last layer is
Softmax classification layer；
S2.3 is trained the convolutional neural networks of design；In the training process, small batch (mini batch) is used every time
Mode will a certain number of silk cocoon training samples be sent into network in, finally using Softmax classification layer export each sample
Prediction result measures the gap between prediction result and legitimate reading by multinomial logic loss function, using the ladder of momentum-type
Degree descent method is that optimization method makes loss function reach minimum, and then realizes the training to convolutional neural networks, is obtained for silkworm
The convolutional neural networks disaggregated model of cocoon classification；
S3 on-line checking process:
S3.1 obtains silk cocoon image online, adjusts industrial camera lens aperture, focal length parameter, light source is installed on the outside of camera,
Silk cocoon image is shot under dark-background；
S3.2 is for shooting silk cocoon image under the dark-background of acquisition, and using the method for template matching, Primary Location goes out dark back
The position of silk cocoon in silk cocoon image is shot under scape；
S3.3 is extended the center that silk cocoon in silk cocoon image is shot under dark-background, and expanding to specified size is W × H
Screenshot afterwards makes only one silk cocoon on every screenshot be located at the center of image；
The size that S3.4 adjusts every screenshot is fixed size；
S3.5 executes the method for normalizing for subtracting mean value, the image after being normalized to the screenshot after having adjusted size；
Image after normalization is sent into trained CNN model by S3.6, by the propagated forward of model, obtains prediction knot
Fruit is to get the classification information for arriving silk cocoon.
2. the online visible detection method of a kind of silk cocoon based on convolutional neural networks according to claim 1, feature exist
In: this method carries out region selection to collected silk cocoon image, and category constructs silk cocoon data set；Design simultaneously training convolutional
Neural network silk cocoon disaggregated model forms the disaggregated model that accuracy of identification is high and parameter amount is small；Captured in real-time figure in on-line checking
Piece chooses silk cocoon region automatically, is input in disaggregated model and is predicted, obtains classification results；
Establish the specific implementation step of silk cocoon data set are as follows:
Step 1: the acquisition of silk cocoon image
According to silk cocoon recognition detection required precision, industrial camera lens aperture, focal length mechanical parameter are adjusted, obtains silk cocoon figure
Picture；
Step 2: picture size pre-processes
(1) gaussian filtering process is carried out in the original image of acquisition, and binaryzation is executed to filtered image, enhances contrast；
(2) barycentric coodinates (x, y) of silk cocoon region in binary map are sought；
(3) it centered on center of gravity, is extended in original image and intercepts out the region of W*H to remove extra background information, make the region
It can completely include all kinds of silk cocoons, wherein W, H are fixed numbers, and specific value is adjusted according to silk cocoon in image proportion
It is whole；Specific extended method is two coordinates for finding the upper left corner and the lower right corner of target area, wherein the point coordinate in the upper left corner
For (x-W/2, y-H/2), the point coordinate in the lower right corner is (x+W/2, y+H/2)；Screenshot is carried out by the two point coordinates, is gone
Except the silk cocoon image after background, step 2 is executed to every obtained original image, establishes silk cocoon image data set.
3. the online visible detection method of a kind of silk cocoon based on convolutional neural networks according to claim 1, feature exist
In: it designs and silk cocoon disaggregated model of the training based on convolutional neural networks implements step are as follows:
Step 1: carrying out size adjusting and normalized to the sample in silk cocoon data set；
The size that image in silk cocoon data set is adjusted by way of squash, is set as 227*227, carries out picture to image
Plain normalized, concrete operations are that all silk cocoon images of data concentration are sought with R, the mean value in tri- channels G, B, then every
The mean value that tri- channels RGB of image subtract corresponding channel realizes normalization；
Step 2: design convolutional neural networks structure；
The convolutional neural networks that this method is built use 8 layers of structure, and first four layers are alternately arranged for convolutional layer with pond layer, layer 5
It is convolutional layer with layer 6, layer 7 is global average pond layer, and the 8th layer is classification layer, and every layer of weight uses Gaussian Profile
Random number realize initialization；In the silk cocoon convolutional neural networks structure of this method, Conv1 layers, Conv2 layers, Conv3 layers,
Conv4 layers are convolutional layer, and every layer of convolution kernel size is 3x3, and the quantity of convolution kernel is respectively 64,128,192,192,
Nonlinear Processing is executed using ReLU function to obtained characteristic pattern after each convolution operation, Pool1 layers, Pool2 layers are most
Great Chiization layer, GlobalAVE Pool are global average pond layer, and the last layer is Softmax classification layer；
Step 3: the convolutional neural networks to design are trained, the convolutional neural networks classification mould for silk cocoon classification is generated
Type；
(1) in the training process, a certain number of silk cocoon training samples are sent into network by the way of small batch every time, most
The prediction result of each sample is exported using Softmax classification layer eventually；
(2) gap between prediction result and legitimate reading is measured by multinomial logic loss function:
Wherein, E is error loss, wherein pnIndicate that the silk cocoon image n of prediction is the probability of its true tag label, N is this
Batch is input to the sum of the silk cocoon image in model；
(3) gradient descent method of momentum-type is used so that loss function is reached minimum for optimization method, wherein momentum parameter is set as
0.9, and then realize the training to convolutional neural networks, obtain the convolutional neural networks disaggregated model for silk cocoon classification.
4. the online visible detection method of a kind of silk cocoon based on convolutional neural networks according to claim 1, feature exist
In: on-line checking process implements step are as follows:
Step 1: Primary Location goes out silk cocoon position in image；
The requirement of real-time online detection accuracy is identified according to silk cocoon, is adjusted industrial camera lens aperture, focal length parameter, is obtained in real time
Silk cocoon image, the overall profile of the qualified cocoon of interception is as template under the present conditions, the Primary Location by way of template matching
Each silk cocoon position out；
Step 2: extending matching position to specified size；
By obtained silk cocoon preliminary matches information, the center (x of each silk cocoon is calculated1,y1), (x2,y2)……(xj,
yj) and intercept on the basis of center the region of W*H around, j indicates the number of silk cocoon, and W is the width of interception area, H
For the height of interception area；The region method of specific interception W*H is two points for finding the upper left corner and the lower right corner of target area
Coordinate, wherein the point coordinate in the upper left corner is (x-W/2, y-H/2), and the point coordinate in the lower right corner is (x+W/2, y+H/2), passes through upper left
Angle and two, lower right corner coordinate carry out screenshot；
Step 3: image normalization handles and is sent into model；
Truncated picture is adjusted to 227*227 size, then the image subtract the normalized of mean value, by what is handled well
Image is sent into trained model, by propagated forward, output category information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810820347.9A CN109215015A (en) | 2018-07-24 | 2018-07-24 | A kind of online visible detection method of silk cocoon based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810820347.9A CN109215015A (en) | 2018-07-24 | 2018-07-24 | A kind of online visible detection method of silk cocoon based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109215015A true CN109215015A (en) | 2019-01-15 |
Family
ID=64990606
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810820347.9A Pending CN109215015A (en) | 2018-07-24 | 2018-07-24 | A kind of online visible detection method of silk cocoon based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109215015A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109870144A (en) * | 2019-02-27 | 2019-06-11 | 山东明佳科技有限公司 | A kind of high-speed, high precision vision positioning system and method |
| CN110503645A (en) * | 2019-08-29 | 2019-11-26 | 国合通用(青岛)测试评价有限公司 | The method that metallograph grain size is determined based on convolutional neural networks |
| CN110567888A (en) * | 2019-09-12 | 2019-12-13 | 中国计量大学 | silkworm cocoon dry shell amount nondestructive testing method based on visible/near infrared spectrum and deep learning |
| CN111146108A (en) * | 2020-01-10 | 2020-05-12 | 上海应用技术大学 | Broken grid defect detection device and method for solar cell panel |
| CN111204476A (en) * | 2019-12-25 | 2020-05-29 | 上海航天控制技术研究所 | Vision-touch fusion fine operation method based on reinforcement learning |
| CN111687069A (en) * | 2020-06-01 | 2020-09-22 | 安徽农业大学 | Intelligent pecan shell and kernel sorting machine based on convolutional neural network algorithm |
| CN112308028A (en) * | 2020-11-25 | 2021-02-02 | 四川省农业科学院蚕业研究所 | Intelligent counting method for silkworm larvae and application thereof |
| CN115311505A (en) * | 2022-10-11 | 2022-11-08 | 四川主干信息技术有限公司 | Silkworm cocoon classification method and purchase system based on cloud service big data platform |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106906520A (en) * | 2017-02-16 | 2017-06-30 | 柳州市自动化科学研究所 | The method and its silk cocoon screening system of the inferior silk cocoon of electronic recognition |
| CN107679579A (en) * | 2017-10-17 | 2018-02-09 | 天津工业大学 | Jujube quality method for separating based on deep learning |
| WO2018052587A1 (en) * | 2016-09-14 | 2018-03-22 | Konica Minolta Laboratory U.S.A., Inc. | Method and system for cell image segmentation using multi-stage convolutional neural networks |
| CN108229561A (en) * | 2018-01-03 | 2018-06-29 | 北京先见科技有限公司 | Particle product defect detection method based on deep learning |
-
2018
- 2018-07-24 CN CN201810820347.9A patent/CN109215015A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018052587A1 (en) * | 2016-09-14 | 2018-03-22 | Konica Minolta Laboratory U.S.A., Inc. | Method and system for cell image segmentation using multi-stage convolutional neural networks |
| CN106906520A (en) * | 2017-02-16 | 2017-06-30 | 柳州市自动化科学研究所 | The method and its silk cocoon screening system of the inferior silk cocoon of electronic recognition |
| CN107679579A (en) * | 2017-10-17 | 2018-02-09 | 天津工业大学 | Jujube quality method for separating based on deep learning |
| CN108229561A (en) * | 2018-01-03 | 2018-06-29 | 北京先见科技有限公司 | Particle product defect detection method based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| 叶子等: "卷积神经网络模型压缩在图像分类中的应用", 《北京信息科技大学学报》 * |
| 汪成龙等: "基于深度学习的马铃薯畸形检测方法研究1", 《惠州学院学报（自然科学版）》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109870144A (en) * | 2019-02-27 | 2019-06-11 | 山东明佳科技有限公司 | A kind of high-speed, high precision vision positioning system and method |
| CN110503645A (en) * | 2019-08-29 | 2019-11-26 | 国合通用(青岛)测试评价有限公司 | The method that metallograph grain size is determined based on convolutional neural networks |
| CN110567888A (en) * | 2019-09-12 | 2019-12-13 | 中国计量大学 | silkworm cocoon dry shell amount nondestructive testing method based on visible/near infrared spectrum and deep learning |
| CN111204476A (en) * | 2019-12-25 | 2020-05-29 | 上海航天控制技术研究所 | Vision-touch fusion fine operation method based on reinforcement learning |
| CN111204476B (en) * | 2019-12-25 | 2021-10-29 | 上海航天控制技术研究所 | Vision-touch fusion fine operation method based on reinforcement learning |
| CN111146108A (en) * | 2020-01-10 | 2020-05-12 | 上海应用技术大学 | Broken grid defect detection device and method for solar cell panel |
| CN111687069A (en) * | 2020-06-01 | 2020-09-22 | 安徽农业大学 | Intelligent pecan shell and kernel sorting machine based on convolutional neural network algorithm |
| CN111687069B (en) * | 2020-06-01 | 2023-02-28 | 安徽农业大学 | Intelligent pecan shell and kernel sorting machine based on convolutional neural network algorithm |
| CN112308028A (en) * | 2020-11-25 | 2021-02-02 | 四川省农业科学院蚕业研究所 | Intelligent counting method for silkworm larvae and application thereof |
| CN112308028B (en) * | 2020-11-25 | 2023-07-14 | 四川省农业科学院蚕业研究所 | Intelligent silkworm larva counting method |
| CN115311505A (en) * | 2022-10-11 | 2022-11-08 | 四川主干信息技术有限公司 | Silkworm cocoon classification method and purchase system based on cloud service big data platform |
| CN115311505B (en) * | 2022-10-11 | 2022-12-27 | 四川主干信息技术有限公司 | Silkworm cocoon classification method and purchase system based on cloud service big data platform |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109215015A (en) | A kind of online visible detection method of silk cocoon based on convolutional neural networks | |
| CN110543878B (en) | Pointer instrument reading identification method based on neural network | |
| CN111179251B (en) | Defect detection system and method based on twin neural network and by utilizing template comparison | |
| He et al. | A fully convolutional neural network for wood defect location and identification | |
| CN106875373B (en) | Mobile phone screen MURA defect detection method based on convolutional neural network pruning algorithm | |
| CN108765412B (en) | Strip steel surface defect classification method | |
| CN110276386A (en) | A kind of apple grading method and system based on machine vision | |
| CN108648168A (en) | IC wafer surface defects detection methods | |
| CN109977808A (en) | A kind of wafer surface defects mode detection and analysis method | |
| CN108520274A (en) | High reflecting surface defect inspection method based on image procossing and neural network classification | |
| CN110148162A (en) | A kind of heterologous image matching method based on composition operators | |
| CN111815564B (en) | Method and device for detecting silk ingots and silk ingot sorting system | |
| CN108985337A (en) | A kind of product surface scratch detection method based on picture depth study | |
| Zhang et al. | An intelligent model for detecting and classifying color-textured fabric defects using genetic algorithms and the Elman neural network | |
| CN103278511B (en) | Wafer defect detection method based on multi-scale corner feature extraction | |
| CN105976356B (en) | A kind of Robust digital figure correlation technique based on joint entropy criterion | |
| CN114119526A (en) | Steel plate surface defect detection and identification system and method based on machine vision | |
| Gao et al. | Detection of wheat unsound kernels based on improved ResNet | |
| Wang et al. | A fabric defect detection system based improved yolov5 detector | |
| Wen et al. | Multi-scene citrus detection based on multi-task deep learning network | |
| Lee et al. | Intelligent classification methods of grain kernels using computer vision analysis | |
| CN109815957A (en) | A kind of character recognition method based on color image under complex background | |
| CN116843615B (en) | Lead frame intelligent total inspection method based on flexible light path | |
| Aferi et al. | Cotton texture segmentation based on image texture analysis using gray level co-occurrence matrix (GLCM) and Euclidean distance | |
| CN116051808A (en) | YOLOv 5-based lightweight part identification and positioning 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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190115 |