CN108389194A - The image-recognizing method and device of denim water washing effect evaluation based on artificial intelligence - Google Patents

The image-recognizing method and device of denim water washing effect evaluation based on artificial intelligence Download PDF

Info

Publication number
CN108389194A
CN108389194A CN201810157496.1A CN201810157496A CN108389194A CN 108389194 A CN108389194 A CN 108389194A CN 201810157496 A CN201810157496 A CN 201810157496A CN 108389194 A CN108389194 A CN 108389194A
Authority
CN
China
Prior art keywords
image
denim
artificial intelligence
sample image
fabric
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.)
Granted
Application number
CN201810157496.1A
Other languages
Chinese (zh)
Other versions
CN108389194B (en
Inventor
余方政
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangzhou Da Long Biotechnology Co Ltd
Original Assignee
Guangzhou Da Long Biotechnology Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guangzhou Da Long Biotechnology Co Ltd filed Critical Guangzhou Da Long Biotechnology Co Ltd
Priority to CN201810157496.1A priority Critical patent/CN108389194B/en
Publication of CN108389194A publication Critical patent/CN108389194A/en
Application granted granted Critical
Publication of CN108389194B publication Critical patent/CN108389194B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/0002Inspection of images, e.g. flaw detection
    • G06T7/0004Industrial image inspection
    • G06T7/0006Industrial image inspection using a design-rule based approach
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20081Training; Learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20084Artificial neural networks [ANN]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30108Industrial image inspection
    • G06T2207/30124Fabrics; Textile; Paper

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)
  • Image Processing (AREA)

Abstract

The image-recognizing method for the denim water washing effect evaluation based on artificial intelligence that this application discloses a kind of.This method includes:Capturing sample image;The sample image is inputted into convolutional neural networks, extracts the image features of the sample image;It trains to obtain discrimination model by the image features of the sample image;And after the verification discrimination model, determine the defeathering effect of fabric.Present application addresses determine the relatively low technical problem of defeathering effect timeliness rate.The defeathering effect recognition result after fabric wash water can be obtained using artificial intelligence machine study by the application, avoid complex characteristic from extracting, while solving the differentiation limitation that human eye is difficult to differentiate.

Description

The image-recognizing method and device of denim water washing effect evaluation based on artificial intelligence
Technical field
This application involves field of image recognition, in particular to a kind of denim water washing effect based on artificial intelligence The image-recognizing method and device of evaluation.
Background technology
Defeathering is needed after gigging, differentiates that the defeathering effect of fabric usually wants professional by virtue of experience to detect by an unaided eye To be judged.
Inventor find, when determining fabric defeathering degree heavy workload, speed is slow, human intervention is serious, result is inaccurate Really.
For the relatively low problem of defeathering effect timeliness rate is determined in the related technology, effective solution side is not yet proposed at present Case.
Invention content
The main purpose of the application is to provide a kind of image knowledge of the denim water washing effect evaluation based on artificial intelligence Other method, to solve the problems, such as to determine that defeathering effect timeliness rate is relatively low.
To achieve the goals above, according to the one side of the application, a kind of denim based on artificial intelligence is provided The image-recognizing method of water washing effect evaluation, the defeathering effect for differentiating fabric.
Image-recognizing method according to denim water washing effect evaluation of the application based on artificial intelligence includes:Acquire sample This image;The sample image is inputted into convolutional neural networks, extracts the image features of the sample image;By described The image features of sample image train to obtain discrimination model;And after the verification discrimination model, determine removing for fabric Capillary effect fruit.
Further, capturing sample image includes:Fabric the first key position image before acquisition wash water;It is knitted after acquisition wash water Object the second key position image;Wherein, the first key position image and the second key position image are same position Image.
Further, the sample image is inputted into convolutional neural networks, extracts the characteristics of image ginseng of the sample image Number includes:The sample image obtains the first enhancing image by image enhancement processing;Figure is executed to the first enhancing image As target classification operation, first is obtained through Yarn image and the first weft yarn image;According to described first through Yarn image and the first weft yarn Image recognition goes out the image features of the sample image.
Further, train to obtain discrimination model by the image features of the sample image include:Using Dropout solves overfitting problem, by a data output node random drop part for a certain layer of neural network when training, improves Model generalization;Learning method selects Adam adaptive approach;Activation primitive selects ReLU nonlinear functions.
To achieve the goals above, according to the another aspect of the application, a kind of denim based on artificial intelligence is provided The pattern recognition device of water washing effect evaluation, the defeathering effect for differentiating fabric.
Pattern recognition device according to denim water washing effect evaluation of the application based on artificial intelligence includes:Acquisition is single Member is used for capturing sample image;Recognition unit extracts the sample for the sample image to be inputted convolutional neural networks The image features of image;Training unit differentiates mould for training to obtain by the image features of the sample image Type;Verification judgement unit determines the defeathering effect of fabric after verifying the discrimination model.
Further, the collecting unit includes:First image acquisition units, the second image acquisition units, described first Image acquisition units, for acquiring fabric the first key position image before wash water;Second image acquisition units, for acquiring The second key position of fabric image after wash water;Wherein, the first key position image is with the second key position image The image of same position.
Further, the recognition unit includes:Enhance elementary area, separate picture unit, feature extraction unit, enhancing Elementary area, for sample image to be obtained the first enhancing image by image enhancement processing;Separate picture unit, for institute It states the first enhancing image and executes image object sort operation, obtain first through Yarn image and the first weft yarn image;Feature extraction list Member, the image features for going out the sample image through Yarn image and the first weft yarn image recognition according to described first.
Further, the training unit includes:Model generalization is improved using Dropout, is solved using Adam methods Tuning parameter problem, learning rate are set as 0.00001, and activation primitive uses ReLU functions, convolutional neural networks structure to have 5 training Layer, first 3 layers are convolutional layer, and latter 2 layers are full articulamentum, last layer is Softmax layers.
Further, the judgement unit includes:Using the network structure trained, defeathering effect is carried out to unknown images Differentiate.
Further, the collecting unit includes:Digit microscope, the fabric include:Denim.
In the embodiment of the present application, convolutional neural networks are inputted using by the sample image, extracts the sample image Image features mode, train to obtain discrimination model by the image features of the sample image, reached and sentenced The purpose for not going out the defeathering effect of fabric to realize accurate, the efficient technique effect for differentiating fabric defeathering effect, and then solves The relatively low technical problem of determining defeathering effect timeliness rate of having determined.
Description of the drawings
The attached drawing constituted part of this application is used for providing further understanding of the present application so that the application's is other Feature, objects and advantages become more apparent upon.The illustrative examples attached drawing and its explanation of the application is for explaining the application, not Constitute the improper restriction to the application.In the accompanying drawings:
Fig. 1 is the image-recognizing method schematic diagram according to the application first embodiment;
Fig. 2 is the image-recognizing method schematic diagram according to the application second embodiment;
Fig. 3 is the image-recognizing method schematic diagram according to the application 3rd embodiment;
Fig. 4 is the pattern recognition device schematic diagram according to the application first embodiment;
Fig. 5 is the pattern recognition device schematic diagram according to the application second embodiment;
Fig. 6 is the pattern recognition device schematic diagram according to the application 3rd embodiment;
Fig. 7 (a) is original image schematic diagram in image recognition processes;
Fig. 7 (b) is enhancing image schematic diagram in image recognition processes;
Fig. 7 (c) is in image recognition processes through Yarn image schematic diagram;And
Fig. 7 (d) is weft yarn image schematic diagram in image recognition processes.
Specific implementation mode
In order to make those skilled in the art more fully understand application scheme, below in conjunction in the embodiment of the present application Attached drawing, technical solutions in the embodiments of the present application are clearly and completely described, it is clear that described embodiment is only The embodiment of the application part, instead of all the embodiments.Based on the embodiment in the application, ordinary skill people The every other embodiment that member is obtained without making creative work should all belong to the model of the application protection It encloses.
It should be noted that term " first " in the description and claims of this application and above-mentioned attached drawing, " Two " etc. be for distinguishing similar object, without being used to describe specific sequence or precedence.It should be appreciated that using in this way Data can be interchanged in the appropriate case, so as to embodiments herein described herein.In addition, term " comprising " and " tool Have " and their any deformation, it is intended that cover it is non-exclusive include, for example, containing series of steps or unit Process, method, system, product or equipment those of are not necessarily limited to clearly to list step or unit, but may include without clear It is listing to Chu or for these processes, method, product or equipment intrinsic other steps or unit.
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase Mutually combination.The application is described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
Fig. 1 is the image recognition according to the denim water washing effect evaluation based on artificial intelligence of the application first embodiment Method schematic diagram;It includes the following steps S102 to step S108:
Step S102, capturing sample image;
Using the collected textile image of electron microscope as sample image.
The quantity of master drawing image can be acquired based on the image of certain amount rank.
For example, by digit microscope respectively to the identical key position of denim is taken pictures before wash water, after wash water, profit With the image features on same position photo before and after image enhancement processing technology extraction wash water, quantitatively sentenced using characteristic parameter The rank of defeathering effect after other wash water.
The sample image is inputted convolutional neural networks by step S104, extracts the characteristics of image ginseng of the sample image Number;
Refer to handling, analyzing and understand picture by computer program by sample image input convolutional neural networks Content automatically extracts characteristics of image by machine learning from picture, completes image recognition.By by the original of a pictures Pixel is constantly abstract, and height is further formed from original image pixels composition point, line foundation structure, then by foundation characteristics such as point, lines Rank feature is finally identified image by the feature learning done in the training process.Machine learning uses a large amount of number According to training, by various arithmetic analysis data, therefrom self study carries out decision and prediction to real problems.Machine learning passes through Neural network carrys out training data model, and most important is exactly to train, and needs hundreds and thousands of or even millions of images to train, directly Weights to the input of neuron are all modulated exactly accurate.
Step S106 trains to obtain discrimination model by the image features of the sample image;
Picture after typing judges the defeathering effect level of fabric automatically by discrimination model comparison.Sentence with needing to analyze Disconnected picture is on the increase so that the processing analysis model in system is more and more, and the depth of machine learning is more and more deeper so that The obtained rank of subsequent processing discriminance analysis is more accurate.
Step S108 after verifying the discrimination model, determines the defeathering effect of fabric.
By the defeathering grade effect discrimination model to deep learning by the training determination of mass data collection, model can be evaluated and tested Accuracy rate on test set.
It can be seen from the above description that the present invention realizes following technique effect:
In the embodiment of the present application, convolutional neural networks are inputted using by the sample image, extracts the sample image Image features mode, train to obtain discrimination model by the image features of the sample image, reached and sentenced The purpose for not going out the defeathering effect of fabric to realize accurate, the efficient technique effect for differentiating fabric defeathering effect, and then solves The relatively low technical problem of determining defeathering effect timeliness rate of having determined.And by big data, artificial intelligence and machine deep learning, The quality for the defeathering effect that fabric is differentiated after wash water is automatically analyzed using image enhancement recognition methods.
According to embodiments of the present invention, as preferred in the present embodiment, as shown in Fig. 2, capturing sample image includes:
Step S202 acquires fabric the first key position image before wash water;
Step S204 acquires fabric the second key position image after wash water;
Step S206, the first key position image and the image that the second key position image is same position.
In the above method, using digit microscope to the identical key position of denim is clapped before wash water and after wash water According to amplification factor is set as 100 times when taking pictures, and camera lens is close to fabrics cloth cover, and database is automatically credited after completion of taking pictures.
According to embodiments of the present invention, it as preferred in the present embodiment, is rolled up as shown in figure 3, the sample image is inputted Product neural network, the image features for extracting the sample image include:
Step S302, the sample image obtain the first enhancing image by image enhancement processing;
The application in convolutional neural networks by automated graphics enhancing, image object separation, image characteristics extraction come Complete image recognition.
Step S304 executes image object sort operation to the first enhancing image, obtains first through Yarn image and the One weft yarn image;
Image object separation is that the image separation of warp thread and weft yarn is realized to washing image, preferably to differentiate defeathering effect Rank.
Step S306 goes out the image spy of the sample image according to described first through Yarn image and the first weft yarn image recognition Levy parameter.
As shown in Fig. 7 (a)-Fig. 7 (d), be illustrated respectively in original image in embodiment, enhancing image, through Yarn image with And the schematic diagram of weft yarn image.
Train to obtain discrimination model by the image features of the sample image include:
As preferred in the present embodiment, in the training process of deep learning network, to prevent convolutional neural networks mistake Fitting problems take certain methods.
For example, solving overfitting problem using some nodes in Dropout random drop networks.
As preferred in the present embodiment, by debugging determining optimum algorithm of multi-layer neural network repeatedly in structure design, this It is since convolutional neural networks are not convex optimization problems, parameter is generally difficult to debug.
For example, mitigating the burden of tuning parameter using Adam adaptive approach.
As preferred in the present embodiment, select correct activation primitive effectively to avoid gradient disperse problem.
For example, using ReLU nonlinear functions as activation primitive, can be very good to transmit gradient, even across plurality of layers Backpropagation, Grad will not substantially reduce.
Following fabric is by taking denim as an example, to the image recognition of the above-mentioned denim water washing effect evaluation based on artificial intelligence Method is further described.
By digit microscope respectively to the identical key position of denim is taken pictures before wash water, after wash water, figure is utilized Image features before and after image intensifying treatment technology extraction wash water on same position photo, are washed using characteristic parameter quantitative identification The rank of defeathering effect after water.
By denim image before and after the washing as sample image, deep learning frame is inputted.Deep learning frame Tensorflow realizes application of the convolutional neural networks image recognition technology in terms of jean water-washing industry defeathering effect differentiation. By early period using the photo that can largely mark off defeathering effect level, certain image processing and analyzing model is established, every time When wash water, respectively to identical key position is taken pictures before wash water, after wash water, the picture of shooting is entered into database, is recorded Picture after entering judges defeathering effect level automatically by the processing analysis model comparison in system;It is analyzed and determined with needs Picture is on the increase so that the processing analysis model in discrimination model is more and more, and the depth of machine learning is more and more deeper so that The obtained rank of subsequent processing discriminance analysis is more accurate.
It should be noted that step shown in the flowchart of the accompanying drawings can be in such as a group of computer-executable instructions It is executed in computer system, although also, logical order is shown in flow charts, and it in some cases, can be with not The sequence being same as herein executes shown or described step.
According to embodiments of the present invention, it additionally provides a kind of for implementing the above-mentioned denim water washing effect based on artificial intelligence The device of the image-recognizing method of evaluation, as shown in figure 4, the device includes:Collecting unit 10 is used for capturing sample image;Know Other unit 20 extracts the image features of the sample image for the sample image to be inputted convolutional neural networks;Instruction Practice unit 30, for training to obtain discrimination model by the image features of the sample image;Judgement unit 40 is verified, is used After verifying the discrimination model, the defeathering effect of fabric is determined.
Using the collected textile image of electron microscope as sample image in the collecting unit 10 of the application.
The quantity of master drawing image can be acquired based on the image of certain amount rank.
For example, by digit microscope respectively to the identical key position of denim is taken pictures before wash water, after wash water, profit With the image features on same position photo before and after image enhancement processing technology extraction wash water, quantitatively sentenced using characteristic parameter The rank of defeathering effect after other wash water.
It refers to by computer journey that the sample image, which is inputted convolutional neural networks, in the recognition unit 20 of the application Sequence handles, analyzes and understand image content, and by machine learning, characteristics of image is automatically extracted from picture, completes image and knows Not.By the way that the original pixels of a pictures are constantly abstract, from original image pixels composition point, line foundation structure, then by point, line Equal foundation characteristics further form high-order feature, are finally known to image by the feature learning done in the training process Not.Machine learning is trained using a large amount of data, and by various arithmetic analysis data, therefrom self study carries out real problems Decision and prediction.By neural network come training data model, most important is exactly to train for machine learning, need it is hundreds and thousands of very It is trained to millions of images, until the weights of the input of neuron are all modulated exactly accurate.
Picture in the training unit 30 of the application after typing judges the defeathering of fabric automatically by discrimination model comparison Effect level.It is on the increase with the picture analyzed and determined is needed so that the processing analysis model in system is more and more, machine The depth of study is more and more deeper so that the obtained rank of subsequent processing discriminance analysis is more accurate.
Pass through the defeathering to deep learning by the training determination of mass data collection in the verification judgement unit 40 of the application Grade effect discrimination model, can evaluate and test accuracy rate of the model on test set.
In the convolutional neural networks of the embodiment of the present application, overfitting problem can be solved using Dropout, it will when training A data output node random drop part for a certain layer of neural network, such as can at random delete the 20% of image point Fall, people and machine are still likely to identify the classification of this pictures at this time.The essence done so be create it is more it is new with Press proof sheet prevents over-fitting by increasing sample size, reduction feature quantity, and it is to spy that Dropout abandons node data every time A kind of sampling of sign.Also many methods can prevent CNN over-fittings, accelerate convergence rate or improve generalization etc..Pass through Model overfitting problem is solved using Dropout.
In the convolutional neural networks of the embodiment of the present application, by debugging repeatedly, selection Adam adaptive approach mitigates The burden of tuning parameter, learning rate are set as 0.00001.The problem of in view of neural network not being often a convex optimization, Many times all there are multiple locally optimal solutions, hardly result in globally optimal solution, selected algorithm may result in result and exist Optimal solution nearby fluctuates, and if learning rate difference may cause neural network to obtain different locally optimal solutions, but The many locally optimal solutions obtained for neural network are also that can reach relatively good classification results, just do not look to passing through nerve Network obtains the optimal solution of problem, and optimal solution is more likely the solution of over-fitting.
It, can be very using ReLU nonlinear functions as activation primitive in the convolutional neural networks of the embodiment of the present application Good transmission gradient, even across the backpropagation of plurality of layers, Grad will not substantially reduce.Since neural network activates letter Number selects improper very likely result in training process gradient disperse problem occur.Non-linear Sigmoid functions are widely used In neural network model, because Sigmoid function output numerical values best suit probability output, and Sigmoid between 0~1 Function central area signal gain is larger, and both sides signal gain is small, and can important feature central area be placed in training, insignificant Feature is placed in two lateral areas.But when the neural network number of plies is more, Sigmoid functions Grad in backpropagation can gradually subtract Small, updating neural network parameter according to the feedback of training data will very slowly.ReLU nonlinear functions perfectly solve The problem of gradient disperse.
According to embodiments of the present invention, as preferred in the present embodiment, as shown in figure 5, the collecting unit 10 includes:The One image acquisition units 101, the second image acquisition units 102, described first image collecting unit 101, before acquiring wash water Fabric the first key position image;Second image acquisition units 102, for acquiring fabric the second key position figure after wash water Picture;Wherein, the first key position image and the image that the second key position image is same position.
Specifically, in above-mentioned collecting unit 10, using digit microscope to the identical pass of denim before wash water and after wash water Key position is taken pictures, and amplification factor is set as 100 times (preferably) when taking pictures, and camera lens is close to fabrics cloth cover, after completion of taking pictures certainly Dynamic deposit database.
According to embodiments of the present invention, as preferred in the present embodiment, as shown in fig. 6, the recognition unit 20 includes:Increase Strong elementary area 201, separate picture unit 202, feature extraction unit 203, enhance elementary area 201, are used for sample image The first enhancing image is obtained by image enhancement processing;Separate picture unit 202, for executing figure to the first enhancing image As target classification operation, first is obtained through Yarn image and the first weft yarn image;Feature extraction unit 203, for according to described the Go out the image features of the sample image once Yarn image and the first weft yarn image recognition.
As shown in Fig. 7 (a)-Fig. 7 (d), be illustrated respectively in original image in embodiment, enhancing image, through Yarn image with And the schematic diagram of weft yarn image.
Image object separation is to realize warp thread and latitude to washing image in the separate picture unit 202 of the embodiment of the present application The image of yarn detaches, preferably to differentiate defeathering effect level.
Following fabric is by taking denim as an example, to the configuration modes of the convolutional neural networks of above-mentioned pattern recognition device into traveling One step explanation.
Jean water-washing industry defeathering effect differentiates in convolutional neural networks structure design, passes through parameter testing repeatedly, net Network structure improves adjustment, and algorithm optimization is final to determine deep learning neural network defeathering degree discrimination model.
Entire defeathering effect discrimination model has 5 layers (not including pond layer and LRN layers) for needing training, and first 3 layers are volume Lamination, latter 2 layers are full articulamentum.Last layer is that the Softmax layers for having 4 classes to export are classified as different defeathering degree, maximum After pond layer appears in the 1st convolutional layer, after the 2nd and the 3rd LRN, after LRN layers there is the 1st maximum pond layer, the 2nd After a and the 3rd convolutional layer.ReLU activation primitives are applied behind 5 layers of trained each layer of layer.Preferably, it is used in training layer Dropout ignores a part of neuron at random, and model is avoided overfitting problem occur.Preferably, optimized by Adam adaptive Method mitigates the burden of parameter testing, and learning rate is set as 0.00001.
Finally, model verifies test set after determining, accuracy rate can reach 96%.The model, which can meet, actually answers With.It is more satisfactory for quick discrimination defeathering degree effect.It solves the problems, such as manually to be difficult to extract validity feature, in very big program It is upper to alleviate dependence of the machine learning model for Feature Engineering.
The foregoing is merely the preferred embodiments of the application, are not intended to limit this application, for the skill of this field For art personnel, the application can have various modifications and variations.Within the spirit and principles of this application, any made by repair Change, equivalent replacement, improvement etc., should be included within the protection domain of the application.

Claims (10)

1. a kind of image-recognizing method of the denim water washing effect evaluation based on artificial intelligence, which is characterized in that for differentiating The defeathering effect of fabric, the method includes:
Capturing sample image;
The sample image is inputted into convolutional neural networks, extracts the image features of the sample image;
It trains to obtain discrimination model by the image features of the sample image;And
After verifying the discrimination model, the defeathering effect of fabric is determined.
2. the image-recognizing method of the denim water washing effect evaluation according to claim 1 based on artificial intelligence, special Sign is that capturing sample image includes:
Fabric the first key position image before acquisition wash water;
Fabric the second key position image after acquisition wash water;
Wherein, the first key position image and the image that the second key position image is same position.
3. the image-recognizing method of the denim water washing effect evaluation according to claim 1 based on artificial intelligence, special Sign is, the sample image is inputted convolutional neural networks, the image features for extracting the sample image include:
The sample image obtains the first enhancing image by image enhancement processing;
Image object sort operation is executed to the first enhancing image, obtains first through Yarn image and the first weft yarn image;
Go out the image features of the sample image through Yarn image and the first weft yarn image recognition according to described first.
4. the image-recognizing method of the denim water washing effect evaluation according to claim 1 based on artificial intelligence, special Sign is, train to obtain discrimination model by the image features of the sample image include:It solved to intend using Dropout Conjunction problem improves model generalization by a data output node random drop part for a certain layer of neural network when training;Study Method selection Adam adaptive approach;Activation primitive selects ReLU nonlinear functions.
5. a kind of pattern recognition device of the denim water washing effect evaluation based on artificial intelligence, which is characterized in that for differentiating The defeathering effect of fabric, described device include:
Collecting unit is used for capturing sample image;
Recognition unit extracts the characteristics of image ginseng of the sample image for the sample image to be inputted convolutional neural networks Number;
Training unit, for training to obtain discrimination model by the image features of the sample image;
Verification judgement unit determines the defeathering effect of fabric after verifying the discrimination model.
6. the pattern recognition device of the denim water washing effect evaluation according to claim 5 based on artificial intelligence, special Sign is that the collecting unit includes:First image acquisition units, the second image acquisition units,
Described first image collecting unit, for acquiring fabric the first key position image before wash water;
Second image acquisition units, for acquiring fabric the second key position image after wash water;
Wherein, the first key position image and the image that the second key position image is same position.
7. the pattern recognition device of the denim water washing effect evaluation according to claim 5 based on artificial intelligence, special Sign is that the recognition unit includes:Enhance elementary area, separate picture unit, feature extraction unit,
Enhance elementary area, for sample image to be obtained the first enhancing image by image enhancement processing;
Separate picture unit obtains first through Yarn image for executing image object sort operation to the first enhancing image With the first weft yarn image;
Feature extraction unit, the figure for going out the sample image through Yarn image and the first weft yarn image recognition according to described first As characteristic parameter.
8. the pattern recognition device of the denim water washing effect evaluation according to claim 5 based on artificial intelligence, special Sign is that the training unit includes:Model generalization is improved using Dropout, solving tuning parameter using Adam methods asks Topic, learning rate are set as 0.00001, and activation primitive uses ReLU functions, convolutional neural networks structure to have 5 trained layers, and first 3 layers are Convolutional layer, latter 2 layers are full articulamentum, last layer is Softmax layers.
9. the pattern recognition device of the denim water washing effect evaluation according to claim 5 based on artificial intelligence, special Sign is that the judgement unit includes:Using the network structure trained, defeathering effect differentiation is carried out to unknown images.
10. the pattern recognition device of the denim water washing effect evaluation according to claim 5 based on artificial intelligence, special Sign is that the collecting unit includes:Digit microscope, the fabric include:Denim.
CN201810157496.1A 2018-02-24 2018-02-24 Image recognition method and device based on artificial intelligence jean washing effect evaluation Active CN108389194B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810157496.1A CN108389194B (en) 2018-02-24 2018-02-24 Image recognition method and device based on artificial intelligence jean washing effect evaluation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810157496.1A CN108389194B (en) 2018-02-24 2018-02-24 Image recognition method and device based on artificial intelligence jean washing effect evaluation

Publications (2)

Publication Number Publication Date
CN108389194A true CN108389194A (en) 2018-08-10
CN108389194B CN108389194B (en) 2020-11-17

Family

ID=63069201

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810157496.1A Active CN108389194B (en) 2018-02-24 2018-02-24 Image recognition method and device based on artificial intelligence jean washing effect evaluation

Country Status (1)

Country Link
CN (1) CN108389194B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109885910A (en) * 2019-01-31 2019-06-14 东华大学 A kind of cowboy's embroidery pattern rinsing colour fading technological parameter modeling method
WO2020113694A1 (en) * 2018-12-04 2020-06-11 深圳码隆科技有限公司 Spinning machine fault monitoring method and apparatus, and server
CN112767062A (en) * 2019-10-21 2021-05-07 阿里巴巴集团控股有限公司 Order generation method, data processing method, equipment, system and storage medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105787562A (en) * 2016-03-24 2016-07-20 上海工程技术大学 Yarn dyed fabric texture type recognition method
CN106845556A (en) * 2017-02-09 2017-06-13 东华大学 A kind of fabric defect detection method based on convolutional neural networks
US9697462B1 (en) * 2012-06-27 2017-07-04 Hrl Laboratories, Llc Synaptic time multiplexing

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9697462B1 (en) * 2012-06-27 2017-07-04 Hrl Laboratories, Llc Synaptic time multiplexing
CN105787562A (en) * 2016-03-24 2016-07-20 上海工程技术大学 Yarn dyed fabric texture type recognition method
CN106845556A (en) * 2017-02-09 2017-06-13 东华大学 A kind of fabric defect detection method based on convolutional neural networks

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
南通大学教务处: "《南通大学优秀毕业设计(论文)集2015届》", 30 April 2016 *
林明旺: "基于卷积神经网络的鱼类图像识别与分类", 《图像与多媒体技术》 *
王静: "牛仔织物水冼色差检测装置及技术", 《中国优秀硕士论文全文数据库》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020113694A1 (en) * 2018-12-04 2020-06-11 深圳码隆科技有限公司 Spinning machine fault monitoring method and apparatus, and server
CN109885910A (en) * 2019-01-31 2019-06-14 东华大学 A kind of cowboy's embroidery pattern rinsing colour fading technological parameter modeling method
CN112767062A (en) * 2019-10-21 2021-05-07 阿里巴巴集团控股有限公司 Order generation method, data processing method, equipment, system and storage medium

Also Published As

Publication number Publication date
CN108389194B (en) 2020-11-17

Similar Documents

Publication Publication Date Title
CN107563431A (en) A kind of image abnormity detection method of combination CNN transfer learnings and SVDD
CN109285139A (en) A kind of x-ray imaging weld inspection method based on deep learning
CN107977671A (en) A kind of tongue picture sorting technique based on multitask convolutional neural networks
CN104766097B (en) Surface of aluminum plate defect classification method based on BP neural network and SVMs
CN108389194A (en) The image-recognizing method and device of denim water washing effect evaluation based on artificial intelligence
CN112417990B (en) Examination student illegal behavior identification method and system
CN111783505A (en) Method and device for identifying forged faces and computer-readable storage medium
CN109858429A (en) A kind of identification of eye fundus image lesion degree and visualization system based on convolutional neural networks
CN109509171A (en) A kind of Fabric Defects Inspection detection method based on GMM and image pyramid
CN108596338A (en) A kind of acquisition methods and its system of neural metwork training collection
CN108710893A (en) A kind of digital image cameras source model sorting technique of feature based fusion
CN107958255A (en) A kind of object detection method and device based on image
CN106250911A (en) A kind of picture classification method based on convolutional neural networks
CN108009592A (en) A kind of diabetic retinal classification of images method
CN109146873A (en) A kind of display screen defect intelligent detecting method and device based on study
CN107958253A (en) A kind of method and apparatus of image recognition
CN110245550A (en) A kind of face noise data collection CNN training method based on overall cosine distribution
CN111583253A (en) Textile crease detection method based on deep learning
CN109800662A (en) A kind of teacher teaches Activity recognition method, system, device and storage medium
CN114112984B (en) Fabric fiber component qualitative method based on self-attention
Haoran et al. Egg crack detection based on support vector machine
CN112927180B (en) Cashmere and wool optical microscope image identification method based on generation of confrontation network
Zhang et al. Fabric defect detection based on visual saliency map and SVM
CN109741351A (en) A kind of classification responsive type edge detection method based on deep learning
CN107545565A (en) A kind of solar energy half tone 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
GR01 Patent grant
GR01 Patent grant