CN105426917A - Component classification method and apparatus - Google Patents
Component classification method and apparatus Download PDFInfo
- Publication number
- CN105426917A CN105426917A CN201510819514.4A CN201510819514A CN105426917A CN 105426917 A CN105426917 A CN 105426917A CN 201510819514 A CN201510819514 A CN 201510819514A CN 105426917 A CN105426917 A CN 105426917A
- Authority
- CN
- China
- Prior art keywords
- drawing picture
- part drawing
- convolutional neural
- neural networks
- classification
- 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
- 238000000034 method Methods 0.000 title claims abstract description 87
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 262
- 238000012549 training Methods 0.000 claims description 159
- 238000012360 testing method Methods 0.000 claims description 42
- 230000008569 process Effects 0.000 abstract description 36
- 230000000694 effects Effects 0.000 abstract description 13
- 238000004364 calculation method Methods 0.000 abstract description 5
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000007781 pre-processing Methods 0.000 description 6
- 238000013529 biological neural network Methods 0.000 description 5
- 238000010276 construction Methods 0.000 description 5
- 238000000605 extraction Methods 0.000 description 5
- 238000013480 data collection Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 238000007639 printing Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
Embodiments of the invention disclose a component classification method and apparatus. The method comprises: inputting a to-be-classified component image into a trained convolutional neural network and calculating an advanced feature of the component image; calculating the probabilities of the component image belonging to the categories by utilizing the advanced feature; and taking the category corresponding to the maximum probability in the probabilities as the category of the component image. As the convolutional neural network can learn the advanced feature of the component image, when the component image is classified by utilizing the convolutional neural network, the collection of the component image is not restricted by scenes, the classification effect is good, and the accuracy is high; and meanwhile, as a local weight of the convolutional neural network is shared, in the process for classifying the component image by utilizing the convolutional neural network, the complexity in calculation can be lowered and the classification efficiency is high.
Description
Technical field
The present invention relates to computer realm, be specifically related to a kind of part classification method and device.
Background technology
Printed-wiring board (PWB) (Printedcircuitboard, be called for short pcb board) refer to as various electronic devices and components provide the circuit board of connection, along with electronic equipment becomes increasingly complex, number of electronic components on pcb board also gets more and more, in order to detect the electronic component on pcb board, need to classify to electronic component, thus automatic marking is carried out to electronic component, to alleviate the workload of artificial plate-making, also for follow-up element testing provides component information.
At present, to be truncated to from pcb board, be positioned at position of components, the part drawing picture that comprises discrete component classify time, mainly based on the feature of traditional machine learning method study image, recycle this feature to classify to part drawing picture, but the feature due to the part drawing picture learnt based on traditional machine learning method is easily subject to the impact of external environment, so in some scenarios, when as even in uneven illumination, it is poor to cause the classifying quality of part drawing picture.
Summary of the invention
Embodiments provide a kind of part classification method and device, to classifying exactly to part drawing picture.
Embodiment of the present invention first aspect provides a kind of part classification method, comprising:
By the convolutional neural networks of part drawing picture to be sorted input after training, and calculate the advanced features of described part drawing picture;
The probability utilizing described advanced features to calculate described part drawing picture to belong to each classification;
Get the classification that classification corresponding to probability maximum in described probability is described part drawing picture.
Embodiment of the present invention second aspect provides a kind of part classification device, comprising:
First computing module, for part drawing picture to be sorted is inputted the convolutional neural networks after training, and calculates the advanced features of described part drawing picture;
Second computing module, the probability utilizing described advanced features to calculate described part drawing picture to belong to each classification;
Sort module, for getting the classification that classification corresponding to probability maximum in described probability is described part drawing picture.
Can find out, in the technical scheme that the embodiment of the present invention provides, by the convolutional neural networks of part drawing picture to be sorted input after training, and calculate the advanced features of described part drawing picture; The probability utilizing described advanced features to calculate described part drawing picture to belong to each classification; Get the classification that classification corresponding to probability maximum in described probability is described part drawing picture.Because convolutional neural networks can learn the advanced features of part drawing picture, when the described embodiment of the present invention utilizes convolutional neural networks to classify to part drawing picture, will the collection of part drawing picture be made not by context restrictions, good classification effect, accuracy be high.
Further, because convolutional neural networks local weight is shared, so utilizing convolutional neural networks in the process of element Images Classification, can reduce computation complexity, classification effectiveness is high.
Accompanying drawing explanation
In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, be briefly described to the accompanying drawing used required in embodiment or description of the prior art below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the prerequisite not paying creative work, other accompanying drawing can also be obtained according to these accompanying drawings.
Fig. 1-a is the schematic flow sheet of a kind of part classification method that first embodiment of the invention provides;
Fig. 1-b is the network structure of convolutional neural networks;
Fig. 1-c is through the network structure of the convolutional neural networks after training;
Fig. 2 is the schematic flow sheet of a kind of part classification method that second embodiment of the invention provides;
Fig. 3 is the structural representation of a kind of part classification device that third embodiment of the invention provides;
Fig. 4 is the structural representation of a kind of part classification device that fourth embodiment of the invention provides;
Fig. 5 is the structural representation of a kind of part classification device that fifth embodiment of the invention provides.
Embodiment
Embodiments provide a kind of part classification method and device, to classifying exactly to part drawing picture.
The present invention program is understood better in order to make those skilled in the art person, below in conjunction with the accompanying drawing in the embodiment of the present invention, technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is only the embodiment of a part of the present invention, instead of whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art, not making the every other embodiment obtained under creative work prerequisite, should belong to the scope of protection of the invention.
Term " first ", " second " and " the 3rd " etc. in instructions of the present invention and claims and above-mentioned accompanying drawing are for distinguishing different object, but not for describing particular order.In addition, term " comprises " and their any distortion, and intention is to cover not exclusive comprising.Such as contain the process of series of steps or unit, method, system, product or equipment and be not defined in the step or unit listed, but also comprise the step or unit do not listed alternatively, or also comprise alternatively for other intrinsic step of these processes, method, product or equipment or unit.
A kind of part classification method of the embodiment of the present invention, a kind of part classification method comprises: by the convolutional neural networks of part drawing picture to be sorted input after training, and calculate the advanced features of described part drawing picture; The probability utilizing described advanced features to calculate described part drawing picture to belong to each classification; Get the classification that classification corresponding to probability maximum in described probability is described part drawing picture.
First be the schematic flow sheet of a kind of part classification method that first embodiment of the invention provides see Fig. 1, Fig. 1.Wherein, as shown in Figure 1, a kind of part classification method that first embodiment of the invention provides can comprise:
S101, by the convolutional neural networks of part drawing picture to be sorted input after training, and calculate the advanced features of described part drawing picture.
Wherein, see Fig. 1-b, Fig. 1-b be the network structure of convolutional neural networks.Convolutional neural networks is a kind of degree of depth learning network, convolutional neural networks has unique superiority with the special construction that its local weight is shared in image procossing, its layout is closer to the biological neural network of reality, weights share the complicacy reducing network, and particularly the image of multidimensional input vector directly can input the complexity that this feature of network avoids data reconstruction in feature extraction and assorting process.When utilizing convolutional neural networks to classify to part drawing picture, first need to utilize a large amount of sample training part classification devices, and then utilize the convolutional neural networks trained to classify to part drawing picture.Because convolutional neural networks can learn part drawing as feature at all levels, comprise low-level feature and advanced features, and the advanced features of image does not affect by photographed scene, even if the part drawing picture also namely taken under the scene of complexity also can utilize advanced features to identify part drawing picture, thus utilize this feature can carry out Classification and Identification to part drawing picture exactly.So, two processes can be divided into, i.e. the advanced features of computing element image during the classification of convolutional neural networks computing element image, and the classification utilizing this advanced features computing element image.
Wherein, part drawing similarly is the image at certain position of components referring to intercept from pcb board, in general, this image is the part drawing picture comprising this element, if but when element missing part, this part drawing picture does not also likely comprise element, or when component card is wrong, this part drawing picture is also likely for comprising the part drawing picture of other element.
Alternatively, in the embodiment of the present invention, the convolutional neural networks of indication is the convolutional neural networks comprising N layer, wherein, N be greater than 1 positive integer.Wherein, the front N-1 layer of this convolutional neural networks is used for computing element image feature at all levels, and the n-th layer of this convolutional neural networks is used for the classification of the feature calculation part drawing picture of the part drawing picture calculated according to front N-1 layer.
Preferably, the value of N is 7.
Preferably, the feature at all levels of part drawing picture that the front N-1 layer of this convolutional neural networks calculates comprises low-level features and advanced features.
S102, the probability utilizing described advanced features to calculate described part drawing picture to belong to each classification.Wherein, the classification of part drawing picture refers to the classification annotation carried out part drawing picture according to the difference of component kind on pcb board, as pcb board having 100 kinds of elements, then the classification of part drawing picture has 100 kinds altogether at least, numeral between available 1-100 carries out class indication to it, also can carry out class indication with other symbol to different elements image category.
Alternatively, in an embodiment of the present invention, the classification of part drawing picture to be sorted can be judged by the probable value of each part drawing picture.
Preferably, for the convolutional neural networks having N layer network, if the front N-1 layer of convolutional neural networks is used for the feature at all levels of computing element image, the feature (comprising low-level features and advanced features) that then n-th layer of this convolutional neural networks is used for calculating according to front N-1 layer calculates the probability that this part drawing picture belongs to each classification, recycles the classification of this probabilistic determination part drawing picture.
S103, get the classification that classification corresponding to probability maximum in the probability of each classification described is described part drawing picture.
Wherein, the probable value that part drawing picture belongs to each classification illustrates the possibility that this part drawing picture belongs to each classification, obviously, probability is larger, it is larger that expression belongs to such other possibility, thus get the classification that classification corresponding to probability maximum in the probability of each classification is part drawing picture, classification results will be made the most accurate.
Alternatively, in possible embodiments more of the present invention, when the element in part drawing picture does not exist, the classification of part drawing picture will be other, if as pcb board there being 100 elements, if classify with the part drawing picture that 1-100 is corresponding to these 100 elements, for the part drawing picture of missing part, classification can be 101.
Can find out, in the scheme of the present embodiment, by the convolutional neural networks of part drawing picture to be sorted input after training, and calculate the advanced features of described part drawing picture; The probability utilizing described advanced features to calculate described part drawing picture to belong to each classification; Get the classification that classification corresponding to probability maximum in described probability is described part drawing picture.Because convolutional neural networks can learn the advanced features of part drawing picture, when the described embodiment of the present invention utilizes convolutional neural networks to classify to part drawing picture, will the collection of part drawing picture be made not by context restrictions, good classification effect, accuracy be high.
Further, because convolutional neural networks local weight is shared, so utilizing convolutional neural networks in the process of element Images Classification, can reduce computation complexity, classification effectiveness is high.
Alternatively, in possible embodiments more of the present invention, described by part drawing picture input through training after convolutional neural networks before, described method also comprises:
Template matches is utilized to obtain the position of element in described part drawing picture and align to described part drawing picture;
Described part drawing picture is normalized, performs described step part drawing picture being inputted the convolutional neural networks after training to trigger.
Be appreciated that, because part drawing similarly is intercept from template image the block element image got off, in order to make neural network more accurate to the calculating of part drawing picture, need to make the element in the part drawing picture be truncated to be positioned at the center of image, and the size of image is normalized, like this to ensure the accuracy of subsequent treatment simultaneously.This process is called preprocessing process.
Alternatively, in possible embodiments more of the present invention, also pre-service can not be carried out to part drawing picture.
Alternatively, in possible embodiments more of the present invention, described by part drawing picture input through training after convolutional neural networks before, described method also comprises:
Create the sample set of described part drawing picture;
Utilize convolutional neural networks described in image recognition database pre-training, obtain described convolutional neural networks initial parameter, described image recognition database comprises other natural image various types of collected from nature;
Based on described convolutional neural networks initial parameter, utilize described sample set to train described convolutional neural networks to finely tune described convolutional neural networks initial parameter further, and it is described by the step of the convolutional neural networks of part drawing picture input after training to trigger execution.
Wherein, the sample set of part drawing picture refers to the part drawing picture for training convolutional neural networks collected from pcb board, image recognition database (ImageNet) be existing collect from nature comprise other image basis database various types of, although ImageNet is not electronic component data collection, but it comprises the natural image of 1,500 ten thousand band marks more than 22000 classifications, the general image feature of each level can be learnt out for pre-training convolutional neural networks, obtain good convolutional neural networks initial parameter value.
Alternatively, in possible embodiments more of the present invention, in order to obtain good classifying quality, so the more sample of collection as much as possible is trained convolutional neural networks when training convolutional neural networks, so the sample taken under needing to gather each scene, as the part drawing picture that the pcb board sample taken when light is bad is truncated to, and from different positions or angle shot to sample image, or the sample image photographed under other complex scene.
Be appreciated that, before obtaining sorter, first need to utilize enough sample sets to train convolutional neural networks, and the quantity general finite of the sample set of the part drawing picture collected, so in order to training convolutional neural networks better, existing ImageNet is utilized first to carry out pre-training to convolutional neural networks, obtain the initial parameter value of convolutional neural networks, again based on this initial parameter, utilize the sample set of the part drawing picture collected further to train convolutional neural networks again, thus obtain final sorter.Be through the network structure of the convolutional neural networks after training see Fig. 1-c, Fig. 1-c, wherein, as shown in fig 1-c, after utilizing the sample set of part drawing picture to carry out further training to convolutional neural networks, finely tune the parameter of convolutional neural networks.Fig. 1-b is after mobile learning with the difference of Fig. 1-c, and the nodes of last one deck of convolutional neural networks becomes present N number of node from original 1000 nodes.Wherein, N is the classification number of part drawing picture.
Alternatively, in possible embodiments more of the present invention, when utilizing ImageNet first to carry out pre-training to convolutional neural networks, the nodes of the last one deck of convolutional neural networks is 1000, when utilizing the sample set of part drawing picture to train based on the neural network after pre-training again, the nodes of last for convolutional neural networks one deck is changed into the classification number of element, as element has N class, then change this layer into N number of node.
Alternatively, in possible embodiments more of the present invention, ImageNet also can not be utilized to carry out pre-training to convolutional neural networks, the sample training convolutional neural networks of more part drawing picture can be gathered.
Alternatively, in possible embodiments more of the present invention, the sample set of described part drawing picture comprises:
The training sample set of described part drawing picture and the test sample book collection of described part drawing picture.
Wherein, the training sample set of part drawing picture is at training stage training convolutional neural networks, and the test sample book collection of part drawing picture is the sample set for the classifying quality at the convolutional neural networks of training stage test after training original training to practice.
Alternatively, in possible embodiments more of the present invention, the training sample set of convolutional neural networks and the test sample book collection of convolutional neural networks acquisition method the same, can get from the sample set of collected part drawing picture the test sample book collection that a part makees part drawing picture.
Alternatively, in possible embodiments more of the present invention, in the training stage of convolutional neural networks, if utilize the test sample book collection of convolutional neural networks test the classifying quality obtained not good time, can train further convolutional neural networks.
Be appreciated that, when training convolutional neural networks, utilize the training sample set pair convolutional neural networks of part drawing picture to train respectively, and the training effect utilizing the test sample book set pair part drawing picture of part drawing picture to carry out testing making convolutional neural networks is better.
Alternatively, in possible embodiments more of the present invention, the sample set of the described part drawing picture of described establishment comprises:
Gather printed circuit board image;
With printed circuit board (PCB) template image for reference, described printed circuit board image intercepts part drawing picture and marks to record the classification of described part drawing picture to described part drawing picture;
The sample set of described part drawing picture is gathered from described part drawing picture after mark.
Be appreciated that owing to being need to classify to each element, so when the sample of acquisition elements image, need the sample set of the image intercepting each element above PCB image to carry out classification based training.Further, in order to distinguish each element sample, so need before training to mark to distinguish different elements to each part drawing picture.
Alternatively, in possible embodiments more of the present invention, camera can be set up on a production line, the pcb board card graphic of batch capture different model, and avoid repeating to take a certain pcb board card with board tracking technique.The pcb board card of each like this model all comprises multiple image pattern, certain pcb board card of the corresponding a certain model of each image pattern, thus the part drawing picture like this on the pcb board card got is also from different board, ensures that sample possesses diversity.
Alternatively, in possible embodiments more of the present invention, describedly in described plated circuit, intercept part drawing picture, comprising:
The positional information of the element above printing plate figure picture is utilized automatically to intercept part drawing picture.
Be appreciated that when after the positional information knowing element, then can automatically intercept part drawing picture according to this positional information.
Alternatively, in possible embodiments more of the present invention, the positional information of described acquisition part drawing picture by the positional information of element recorded in board-like file, or can be obtained by the positional information manually marked.
Alternatively, in possible embodiments more of the present invention, describedly mark carried out to part drawing picture comprise:
Mark according to element class information.
Be appreciated that and need the classification to element to distinguish, thus the classification of recording element could carry out missing part detection to element exactly when training.
Alternatively, in other possible embodiments of the present invention, also can be marked element the mode that element is distinguished by other.
Alternatively, in possible embodiments more of the present invention, before the sample set of the described part drawing picture of described collection, described method also comprises:
Template matches is utilized to obtain the position of element in described part drawing picture and align to described part drawing picture;
Described part drawing picture is normalized, to trigger the step of the sample set performing the described part drawing picture of described collection.
Be appreciated that, the process of carrying out with utilizing convolutional neural networks testing is similar, align to make element be positioned at the center of image to image when gathering the sample image of training convolutional neural networks, and image is normalized, this process is called the preprocessing process to image, carries out pre-service training effect can be made better to part drawing picture.
Alternatively, in possible embodiments more of the present invention, when utilizing convolutional neural networks to classify to part drawing picture, if carry out pre-service in the training stage to element image pattern, when convolutional neural networks so after utilizing training is to element image measurement, also need to carry out pre-service to part drawing picture; If do not carry out pre-service to part drawing picture in the training stage, when the convolutional neural networks so after utilizing training is tested part drawing picture, also pre-service is not carried out to part drawing picture.For the ease of better understanding and implement the such scheme of the embodiment of the present invention, the application scenarios concrete below in conjunction with some is illustrated.
Refer to Fig. 2, Fig. 2 is the schematic flow sheet of a kind of part classification method that second embodiment of the invention provides, and wherein, as shown in Figure 2, a kind of part classification method that second embodiment of the invention provides can comprise:
S201, collection printed circuit board image.
Wherein, pcb board card graphic refers to the image comprising part drawing picture that can directly photograph, part drawing similarly is the image at certain position of components referring to intercept from pcb board, in general, this image is the part drawing picture comprising this element, if but when element missing part, this part drawing picture does not also likely comprise element, or when component card is wrong, this part drawing picture also likely comprises the part drawing picture of other element;
The classification of part drawing picture refers to the classification annotation carried out part drawing picture according to the difference of component kind on pcb board, as pcb board having 100 kinds of elements, then the classification of part drawing picture has 100 kinds altogether at least, numeral between available 1-100 carries out class indication to it, also can carry out class indication with other symbol to different elements image category.
Alternatively, in possible embodiments more of the present invention, camera can be set up on a production line, the pcb board card graphic of batch capture different model, and avoid repeating to take a certain pcb board card with board tracking technique.The pcb board card of each like this model all comprises multiple image pattern, certain pcb board card of the corresponding a certain model of each image pattern, thus the part drawing picture like this on the pcb board card got is also from different board, simultaneously, the sample taken under needing to take each scene, as the pcb board image taken when light is bad, and from different positions or angle shot to pcb board image, thus ensure that the element image pattern be truncated to from pcb board image possesses diversity.
Being appreciated that to classify to element, first needing to gather pcb board card graphic, then intercepting part drawing picture from pcb board card graphic.
S202, with printed circuit board (PCB) template image for reference, printed circuit board image intercepts part drawing picture and part drawing picture is marked with the classification of recording element image.
Alternatively, in possible embodiments more of the present invention, pcb board card graphic intercepts the sample image of part drawing picture as part drawing picture, for training convolutional neural networks, also can as the test pattern of part drawing picture, for classifying to test pattern based on the convolutional neural networks trained.
Being appreciated that when intercepting the sample of part drawing picture, owing to being need to classify to each element, so when the sample of acquisition elements image, needing the sample set of the image intercepting each element above PCB image to carry out classification based training.Further, in order to distinguish each element sample, so need before training to mark to distinguish different elements to each part drawing picture.
Alternatively, in possible embodiments more of the present invention, describedly in described plated circuit, intercept part drawing picture, comprising:
The positional information of the element above printing plate figure picture is utilized automatically to intercept part drawing picture.
Be appreciated that when after the positional information knowing element, then can automatically intercept part drawing picture according to this positional information.
Alternatively, in possible embodiments more of the present invention, the positional information of described acquisition part drawing picture by the positional information of element recorded in board-like file, or can be obtained by the positional information manually marked.
Alternatively, in possible embodiments more of the present invention, describedly mark carried out to part drawing picture comprise:
Mark according to element class information.
Be appreciated that and need the classification to element to distinguish, thus the classification of recording element could carry out missing part detection to element exactly when training.
Alternatively, in other possible embodiments of the present invention, also can be marked element the mode that element is distinguished by other.
S203, from the part drawing picture after mark the sample set of acquisition elements image.
Wherein, the sample set of part drawing picture refers to the part drawing picture for training convolutional neural networks collected from pcb board.
Alternatively, in possible embodiments more of the present invention, before the sample set of the described part drawing picture of described collection, described method also comprises:
Template matches is utilized to obtain the position of element in described part drawing picture and align to described part drawing picture;
Described part drawing picture is normalized, to trigger the step of the sample set performing the described part drawing picture of described collection.
Be appreciated that, the process of carrying out with utilizing convolutional neural networks testing is similar, align to make element be positioned at the center of image to image when gathering the sample image of training convolutional neural networks, and image is normalized, this process is called the preprocessing process to image, carries out pre-service training effect can be made better to part drawing picture.
Alternatively, in possible embodiments more of the present invention, in order to obtain good classifying quality, so the more sample of collection as much as possible is trained convolutional neural networks when training convolutional neural networks, sample set due to part drawing picture refers to the part drawing picture for training convolutional neural networks collected from pcb board, so be appreciated that diversity when ensureing pcb board collection, the diversity of part drawing picture will be ensured.
Alternatively, in possible embodiments more of the present invention, the sample set of described part drawing picture comprises:
The training sample set of described part drawing picture and the test sample book collection of described part drawing picture.
Wherein, the training sample set of part drawing picture is at training stage training convolutional neural networks, and the test sample book collection of part drawing picture is the sample set for the classifying quality at the convolutional neural networks of training stage test after training original training to practice.
Alternatively, in possible embodiments more of the present invention, the training sample set of convolutional neural networks and the test sample book collection of convolutional neural networks acquisition method the same, can get from the sample set of collected part drawing picture the test sample book collection that a part makees part drawing picture.
Alternatively, in possible embodiments more of the present invention, in the training stage of convolutional neural networks, if utilize the test sample book collection of convolutional neural networks test the classifying quality obtained not good time, can train further convolutional neural networks.
Be appreciated that, when training convolutional neural networks, utilize the training sample set pair convolutional neural networks of part drawing picture to train respectively, and the training effect utilizing the test sample book set pair part drawing picture of part drawing picture to carry out testing making convolutional neural networks is better.S204, utilize image recognition database pre-training convolutional neural networks, obtain convolutional neural networks initial parameter.
Wherein, described image recognition database comprises other natural image various types of collected from nature, image recognition database (ImageNet) be existing collect from nature comprise other image basis database various types of, although ImageNet is not electronic component data collection, but it comprises the natural image of 1,500 ten thousand band marks more than 22000 classifications, the general image feature of each level can be learnt out for pre-training convolutional neural networks, obtain good convolutional neural networks initial parameter value.
Be appreciated that, before obtaining sorter, first need to utilize enough sample sets to train convolutional neural networks, and the quantity general finite of the sample set of the part drawing picture collected, so in order to training convolutional neural networks better, existing ImageNet is utilized first to carry out pre-training to convolutional neural networks, obtain the initial parameter value of convolutional neural networks, again based on this initial parameter, utilize the sample set of the part drawing picture collected further to train convolutional neural networks again, thus obtain final sorter.
Alternatively, in possible embodiments more of the present invention, also can not utilize image recognition database pre-training convolutional neural networks, so now can be used for training convolutional neural networks by acquisition elements image pattern as much as possible.
S205, based on convolutional neural networks initial parameter, utilize the further training convolutional neural networks of sample set to finely tune convolutional neural networks initial parameter.
Wherein, convolutional neural networks is a kind of degree of depth learning network, convolutional neural networks has unique superiority with the special construction that its local weight is shared in image procossing, its layout is closer to the biological neural network of reality, weights share the complicacy reducing network, and particularly the image of multidimensional input vector directly can input the complexity that this feature of network avoids data reconstruction in feature extraction and assorting process.When utilizing convolutional neural networks to classify to part drawing picture, first need to utilize a large amount of sample training part classification devices, and then utilize the convolutional neural networks trained to classify to part drawing picture.Because convolutional neural networks can learn part drawing as feature at all levels, comprise low-level feature and advanced features, and the advanced features of image does not affect by photographed scene, even if the part drawing picture also namely taken under the scene of complexity also can utilize advanced features to identify part drawing picture, thus utilize this feature can carry out Classification and Identification to part drawing picture exactly.
Alternatively, in possible embodiments more of the present invention, when utilizing ImageNet first to carry out pre-training to convolutional neural networks, the nodes of the last one deck of convolutional neural networks is 1000, when utilizing the sample set of part drawing picture to train based on the neural network after pre-training again, the nodes of last for convolutional neural networks one deck is changed into the classification number of element, as element has N class, then change this layer into N number of node.
Alternatively, in possible embodiments more of the present invention, also based on the initial parameter of the convolutional neural networks after ImageNet training, but can not directly utilize more element image pattern training convolutional neural networks.
S206, by the convolutional neural networks of part drawing picture to be sorted input after training, and calculate the advanced features of described part drawing picture.
Alternatively, in the embodiment of the present invention, the convolutional neural networks of indication is the convolutional neural networks comprising N layer, wherein, N be greater than 1 positive integer.Wherein, the front N-1 layer of this convolutional neural networks is used for computing element image feature at all levels, and the n-th layer of this convolutional neural networks is used for the classification of the feature calculation part drawing picture of the part drawing picture calculated according to front N-1 layer.
Preferably, the value of N is 7.
Preferably, the feature at all levels of part drawing picture that the front N-1 layer of this convolutional neural networks calculates comprises low-level features and advanced features.
S207, advanced features computing element image is utilized to belong to the probability of each classification.
Alternatively, in an embodiment of the present invention, the classification of part drawing picture to be sorted can be judged by the probable value of each part drawing picture.
Preferably, for the convolutional neural networks having N layer network, if the front N-1 layer of convolutional neural networks is used for the feature at all levels of computing element image, the feature (comprising low-level features and advanced features) that then n-th layer of this convolutional neural networks is used for calculating according to front N-1 layer calculates the probability that this part drawing picture belongs to each classification, recycles the classification of this probabilistic determination part drawing picture.
Alternatively, in possible embodiments more of the present invention, described by part drawing picture input through training after convolutional neural networks before, described method also comprises:
Template matches is utilized to obtain the position of element in described part drawing picture and align to described part drawing picture;
Described part drawing picture is normalized, performs described step part drawing picture being inputted the convolutional neural networks after training to trigger.
Be appreciated that, because part drawing similarly is intercept from template image the block element image got off, in order to make neural network more accurate to the calculating of part drawing picture, need to make the element in the part drawing picture be truncated to be positioned at the center of image, and the size of image is normalized, like this to ensure the accuracy of subsequent treatment simultaneously.This process is called preprocessing process.
Alternatively, in possible embodiments more of the present invention, also pre-service can not be carried out to part drawing picture.
Alternatively, in possible embodiments more of the present invention, when utilizing convolutional neural networks to classify to part drawing picture, if carry out pre-service in the training stage to element image pattern, when convolutional neural networks so after utilizing training is to element Images Classification, also need to carry out pre-service to part drawing picture; If do not carry out pre-service to part drawing picture in the training stage, when the convolutional neural networks so after utilizing training is classified to part drawing picture, also pre-service is not carried out to part drawing picture.
S208, get the classification that classification corresponding to probability maximum in the probability of each classification is part drawing picture.
Wherein, the probable value that part drawing picture belongs to each classification illustrates the possibility that this part drawing picture belongs to each classification, obviously, probability is larger, it is larger that expression belongs to such other possibility, thus get the classification that classification corresponding to probability maximum in the probability of each classification is part drawing picture, classification results will be made the most accurate.
Alternatively, in possible embodiments more of the present invention, when the element in part drawing picture does not exist, the classification of part drawing picture will be other, if as pcb board there being 100 elements, if classify with the part drawing picture that 1-100 is corresponding to these 100 elements, for the part drawing picture of missing part, classification can be 101.
Can find out, in the scheme of the present embodiment, by the convolutional neural networks of part drawing picture to be sorted input after training, and calculate the advanced features of described part drawing picture; The probability utilizing described advanced features to calculate described part drawing picture to belong to each classification; Get the classification that classification corresponding to probability maximum in described probability is described part drawing picture.Because convolutional neural networks can learn the advanced features of part drawing picture, when the described embodiment of the present invention utilizes convolutional neural networks to classify to part drawing picture, will the collection of part drawing picture be made not by context restrictions, good classification effect, accuracy be high.
Further, because convolutional neural networks local weight is shared, so utilizing convolutional neural networks in the process of element Images Classification, can reduce computation complexity, classification effectiveness is high.
The embodiment of the present invention also provides a kind of part classification device, and this device comprises:
First computing module, for part drawing picture to be sorted is inputted the convolutional neural networks after training, and calculates the advanced features of described part drawing picture;
Second computing module, the probability utilizing described advanced features to calculate described part drawing picture to belong to each classification;
Sort module, for getting the classification that classification corresponding to probability maximum in described probability is described part drawing picture.
Concrete, refer to Fig. 3, Fig. 3 is the structural representation of a kind of part classification device that third embodiment of the invention provides, and wherein, as shown in Figure 3, a kind of part classification device 300 that third embodiment of the invention provides can comprise:
First computing module 310, second computing module 320 and sort module 330.
Wherein, the first computing module 310, for part drawing picture to be sorted is inputted the convolutional neural networks after training, and calculates the advanced features of described part drawing picture.
Wherein, convolutional neural networks is a kind of degree of depth learning network, convolutional neural networks has unique superiority with the special construction that its local weight is shared in image procossing, its layout is closer to the biological neural network of reality, weights share the complicacy reducing network, and particularly the image of multidimensional input vector directly can input the complexity that this feature of network avoids data reconstruction in feature extraction and assorting process.When utilizing convolutional neural networks to classify to part drawing picture, first need to utilize a large amount of sample training part classification devices, and then utilize the convolutional neural networks trained to classify to part drawing picture.Because convolutional neural networks can learn part drawing as feature at all levels, comprise low-level feature and advanced features, and the advanced features of image does not affect by photographed scene, even if the part drawing picture also namely taken under the scene of complexity also can utilize advanced features to identify part drawing picture, thus utilize this feature can carry out Classification and Identification to part drawing picture exactly.So, two processes can be divided into, i.e. the advanced features of computing element image during the classification of convolutional neural networks computing element image, and the classification utilizing this advanced features computing element image.
Wherein, part drawing similarly is the image at certain position of components referring to intercept from pcb board, in general, this image is the part drawing picture comprising this element, if but when element missing part, this part drawing picture does not also likely comprise element, or when component card is wrong, this part drawing picture also likely comprises the part drawing picture of other element.
Alternatively, in the embodiment of the present invention, the convolutional neural networks of indication is the convolutional neural networks comprising N layer, wherein, N be greater than 1 positive integer.Wherein, the front N-1 layer of this convolutional neural networks is used for computing element image feature at all levels, and the n-th layer of this convolutional neural networks is used for the classification of the feature calculation part drawing picture of the part drawing picture calculated according to front N-1 layer.
Preferably, the value of N is 7.
Preferably, the feature at all levels of part drawing picture that the front N-1 layer of this convolutional neural networks calculates comprises low-level features and advanced features.
Second computing module 320, the probability calculating described part drawing picture for utilizing described advanced features and belong to each classification.
Wherein, the classification of part drawing picture refers to the classification annotation carried out part drawing picture according to the difference of component kind on pcb board, as pcb board having 100 kinds of elements, then the classification of part drawing picture has 100 kinds altogether at least, numeral between available 1-100 carries out class indication to it, also can carry out class indication with other symbol to different elements image category.
Alternatively, in an embodiment of the present invention, the classification of part drawing picture to be sorted can be judged by the probable value of each part drawing picture.
Preferably, for the convolutional neural networks having N layer network, if the front N-1 layer of convolutional neural networks is used for the feature at all levels of computing element image, the feature (comprising low-level features and advanced features) that then n-th layer of this convolutional neural networks is used for calculating according to front N-1 layer calculates the probability that this part drawing picture belongs to each classification, recycles the classification of this probabilistic determination part drawing picture.
Sort module 330, for getting the classification that classification corresponding to probability maximum in the probability of each classification described is described part drawing picture.
Wherein, the probable value that part drawing picture belongs to each classification illustrates the possibility that this part drawing picture belongs to each classification, obviously, probability is larger, it is larger that expression belongs to such other possibility, thus get the classification that classification corresponding to probability maximum in the probability of each classification is part drawing picture, classification results will be made the most accurate.
Alternatively, in possible embodiments more of the present invention, when the element in part drawing picture does not exist, the classification of part drawing picture will be other, if as pcb board there being 100 elements, if classify with the part drawing picture that 1-100 is corresponding to these 100 elements, for the part drawing picture of missing part, classification can be 101.
Be understandable that, the function of each functional module of the part classification device 300 of the present embodiment can according to the method specific implementation in said method embodiment, and its specific implementation process with reference to the associated description of said method embodiment, can repeat no more herein.
Can find out, in the scheme of the present embodiment, part classification device 300 by the convolutional neural networks of part drawing picture to be sorted input after training, and calculates the advanced features of described part drawing picture; Part classification device 300 recycles described advanced features and calculates the probability that described part drawing picture belongs to each classification; And get the classification that classification corresponding to probability maximum in described probability is described part drawing picture.Because convolutional neural networks can learn the advanced features of part drawing picture, when the described embodiment of the present invention utilizes convolutional neural networks to classify to part drawing picture, will the collection of part drawing picture be made not by context restrictions, good classification effect, accuracy be high.
Further, because convolutional neural networks local weight is shared, so utilizing convolutional neural networks in the process of element Images Classification, can reduce computation complexity, classification effectiveness is high.
Refer to Fig. 4, Fig. 4 is the structural representation of a kind of part classification device that fourth embodiment of the invention provides, and wherein, as shown in Figure 4, a kind of part classification device 400 that fourth embodiment of the invention provides can comprise:
First computing module 410, second computing module 420 and sort module 430.
Wherein, the first computing module 410, for part drawing picture to be sorted is inputted the convolutional neural networks after training, and calculates the advanced features of described part drawing picture.
Wherein, convolutional neural networks is a kind of degree of depth learning network, convolutional neural networks has unique superiority with the special construction that its local weight is shared in image procossing, its layout is closer to the biological neural network of reality, weights share the complicacy reducing network, and particularly the image of multidimensional input vector directly can input the complexity that this feature of network avoids data reconstruction in feature extraction and assorting process.When utilizing convolutional neural networks to classify to part drawing picture, first need to utilize a large amount of sample training part classification devices, and then utilize the convolutional neural networks trained to classify to part drawing picture.Because convolutional neural networks can learn part drawing as feature at all levels, comprise low-level feature and advanced features, and the advanced features of image does not affect by photographed scene, even if the part drawing picture also namely taken under the scene of complexity also can utilize advanced features to identify part drawing picture, thus utilize this feature can carry out Classification and Identification to part drawing picture exactly.So, two processes can be divided into, i.e. the advanced features of computing element image during the classification of convolutional neural networks computing element image, and the classification utilizing this advanced features computing element image.
Wherein, part drawing similarly is the image at certain position of components referring to intercept from pcb board, in general, this image is the part drawing picture comprising this element, if but when element missing part, this part drawing picture does not also likely comprise element, or when component card is wrong, this part drawing picture also likely comprises the part drawing picture of other element.
Alternatively, in the embodiment of the present invention, the convolutional neural networks of indication is the convolutional neural networks comprising N layer, wherein, N be greater than 1 positive integer.Wherein, the front N-1 layer of this convolutional neural networks is used for computing element image feature at all levels, and the n-th layer of this convolutional neural networks is used for the classification of the feature calculation part drawing picture of the part drawing picture calculated according to front N-1 layer.
Preferably, the value of N is 7.
Preferably, the feature at all levels of part drawing picture that the front N-1 layer of this convolutional neural networks calculates comprises low-level features and advanced features.
Second computing module 420, the probability calculating described part drawing picture for utilizing described advanced features and belong to each classification.
Wherein, the classification of part drawing picture refers to the classification annotation carried out part drawing picture according to the difference of component kind on pcb board, as pcb board having 100 kinds of elements, then the classification of part drawing picture has 100 kinds altogether at least, numeral between available 1-100 carries out class indication to it, also can carry out class indication with other symbol to different elements image category.
Alternatively, in an embodiment of the present invention, the classification of part drawing picture to be sorted can be judged by the probable value of each part drawing picture.
Preferably, for the convolutional neural networks having N layer network, if the front N-1 layer of convolutional neural networks is used for the feature at all levels of computing element image, the feature (comprising low-level features and advanced features) that then n-th layer of this convolutional neural networks is used for calculating according to front N-1 layer calculates the probability that this part drawing picture belongs to each classification, recycles the classification of this probabilistic determination part drawing picture.
Sort module 430, for getting the classification that classification corresponding to probability maximum in the probability of each classification described is described part drawing picture.
Wherein, the probable value that part drawing picture belongs to each classification illustrates the possibility that this part drawing picture belongs to each classification, obviously, probability is larger, it is larger that expression belongs to such other possibility, thus get the classification that classification corresponding to probability maximum in the probability of each classification is part drawing picture, classification results will be made the most accurate.
Alternatively, in possible embodiments more of the present invention, when the element in part drawing picture does not exist, the classification of part drawing picture will be other, if as pcb board there being 100 elements, if classify with the part drawing picture that 1-100 is corresponding to these 100 elements, for the part drawing picture of missing part, classification can be 101.
Alternatively, in possible embodiments more of the present invention, described part classification device 400 also comprises:
Pretreatment module 440, obtains the position of element in described part drawing picture for utilizing template matches and aligns to described part drawing picture;
Described part drawing picture is normalized, performs described by the step of the convolutional neural networks of part drawing picture to be sorted input after training to trigger described first computing module 410.
Be appreciated that, because part drawing similarly is intercept from template image the block element image got off, in order to make neural network more accurate to the calculating of part drawing picture, need to make the element in the part drawing picture be truncated to be positioned at the center of image, and the size of image is normalized, like this to ensure the accuracy of subsequent treatment simultaneously.This process is called preprocessing process.
Alternatively, in possible embodiments more of the present invention, also pre-service can not be carried out to part drawing picture.
Alternatively, in possible embodiments more of the present invention, described device also comprises:
Sample creation module 450, for creating the sample set of described part drawing picture;
First training module 460, for utilizing convolutional neural networks described in image recognition database pre-training, obtains described convolutional neural networks initial parameter, and described image recognition database comprises other natural image various types of collected from nature;
Second training module 470, for based on described convolutional neural networks initial parameter, utilize described sample set to train described convolutional neural networks to finely tune described convolutional neural networks initial parameter further, and trigger described first computing module 410 and perform described by the step of convolutional neural networks of part drawing picture to be sorted input after training.
Wherein, the sample set of part drawing picture refers to the part drawing picture for training convolutional neural networks collected from pcb board, image recognition database (ImageNet) be existing collect from nature comprise other image basis database various types of, although ImageNet is not electronic component data collection, but it comprises the natural image of 1,500 ten thousand band marks more than 22000 classifications, the general image feature of each level can be learnt out for pre-training convolutional neural networks, obtain good convolutional neural networks initial parameter value.
Alternatively, in possible embodiments more of the present invention, in order to obtain good classifying quality, so the more sample of collection as much as possible is trained convolutional neural networks when training convolutional neural networks, so the sample taken under needing to gather each scene, as the part drawing picture that the pcb board sample taken when light is bad is truncated to, and from different positions or angle shot to sample image, or the sample image photographed under other complex scene.
Be appreciated that, before obtaining sorter, first need to utilize enough sample sets to train convolutional neural networks, and the quantity general finite of the sample set of the part drawing picture collected, so in order to training convolutional neural networks better, existing ImageNet is utilized first to carry out pre-training to convolutional neural networks, obtain the initial parameter value of convolutional neural networks, again based on this initial parameter, utilize the sample set of the part drawing picture collected further to train convolutional neural networks again, thus obtain final sorter.
Alternatively, in possible embodiments more of the present invention, when utilizing ImageNet first to carry out pre-training to convolutional neural networks, the nodes of the last one deck of convolutional neural networks is 1000, when utilizing the sample set of part drawing picture to train based on the neural network after pre-training again, the nodes of last for convolutional neural networks one deck is changed into the classification number of element, as element has N class, then change this layer into N number of node.
Alternatively, in possible embodiments more of the present invention, ImageNet also can not be utilized to carry out pre-training to convolutional neural networks, the sample training convolutional neural networks of more part drawing picture can be gathered.
Alternatively, in possible embodiments more of the present invention, the sample set of described part drawing picture comprises:
The training sample set of described part drawing picture and the test sample book collection of described part drawing picture.
Wherein, the training sample set of part drawing picture is at training stage training convolutional neural networks, and the test sample book collection of part drawing picture is the sample set for the classifying quality at the convolutional neural networks of training stage test after training original training to practice.
Alternatively, in possible embodiments more of the present invention, the training sample set of convolutional neural networks and the test sample book collection of convolutional neural networks acquisition method the same, can get from the sample set of collected part drawing picture the test sample book collection that a part makees part drawing picture.
Alternatively, in possible embodiments more of the present invention, in the training stage of convolutional neural networks, if utilize the test sample book collection of convolutional neural networks test the classifying quality obtained not good time, can train further convolutional neural networks.
Be appreciated that, when training convolutional neural networks, utilize the training sample set pair convolutional neural networks of part drawing picture to train respectively, and the training effect utilizing the test sample book set pair part drawing picture of part drawing picture to carry out testing making convolutional neural networks is better.
Alternatively, in possible embodiments more of the present invention, described sample creation module 450 comprises:
First collecting unit 451, for gathering printed circuit board image;
Interception unit 452, for printed circuit board (PCB) template image for reference, described printed circuit board image intercepts part drawing picture and marks to record the classification of described part drawing picture to described part drawing picture;
Second collecting unit 453, for gathering the sample set of described part drawing picture from described part drawing picture after mark.
Be appreciated that owing to being need to classify to each element, so when the sample of acquisition elements image, need the sample set of the image intercepting each element above PCB image to carry out classification based training.Further, in order to distinguish each element sample, so need before training to mark to distinguish different elements to each part drawing picture.
Alternatively, in possible embodiments more of the present invention, camera can be set up on a production line, the pcb board card graphic of batch capture different model, and avoid repeating to take a certain pcb board card with board tracking technique.The pcb board card of each like this model all comprises multiple image pattern, certain pcb board card of the corresponding a certain model of each image pattern, thus the part drawing picture like this on the pcb board card got is also from different board, ensures that sample possesses diversity.
Alternatively, in possible embodiments more of the present invention, described sample creation module 440 intercepts part drawing picture in described plated circuit, comprising:
The positional information of the element above printing plate figure picture is utilized automatically to intercept part drawing picture.
Be appreciated that when after the positional information knowing element, then can automatically intercept part drawing picture according to this positional information.
Alternatively, in possible embodiments more of the present invention, the positional information of described acquisition part drawing picture by the positional information of element recorded in board-like file, or can be obtained by the positional information manually marked.
Alternatively, in possible embodiments more of the present invention, described sample creation module 450 pairs of part drawing pictures carry out mark and comprise:
Mark according to element class information.
Be appreciated that and need the classification to element to distinguish, thus the classification of recording element could carry out missing part detection to element exactly when training.
Alternatively, in other possible embodiments of the present invention, also can be marked element the mode that element is distinguished by other.
Alternatively, in other possible embodiments of the present invention, described pretreatment module 470,
Also for utilizing template matches obtain the position of element in described part drawing picture and align to described part drawing picture;
Described part drawing picture is normalized, to trigger the step that described sample creation module 450 performs the sample set of the described part drawing picture of described collection.
Be appreciated that, the process of carrying out with utilizing convolutional neural networks testing is similar, align to make element be positioned at the center of image to image when gathering the sample image of training convolutional neural networks, and image is normalized, this process is called the preprocessing process to image, carries out pre-service training effect can be made better to part drawing picture.
Alternatively, in possible embodiments more of the present invention, when utilizing convolutional neural networks to classify to part drawing picture, if carry out pre-service in the training stage to element image pattern, when convolutional neural networks so after utilizing training is to element image measurement, also need to carry out pre-service to part drawing picture; If do not carry out pre-service to part drawing picture in the training stage, when the convolutional neural networks so after utilizing training is tested part drawing picture, also pre-service is not carried out to part drawing picture.
Be understandable that, the function of each functional module of the part classification device 400 of the present embodiment can according to the method specific implementation in said method embodiment, and its specific implementation process with reference to the associated description of said method embodiment, can repeat no more herein.
Can find out, in the scheme of the present embodiment, part classification device 400 by the convolutional neural networks of part drawing picture to be sorted input after training, and calculates the advanced features of described part drawing picture; Part classification device 400 recycles described advanced features and calculates the probability that described part drawing picture belongs to each classification; And get the classification that classification corresponding to probability maximum in described probability is described part drawing picture.Because convolutional neural networks can learn the advanced features of part drawing picture, when the described embodiment of the present invention utilizes convolutional neural networks to classify to part drawing picture, will the collection of part drawing picture be made not by context restrictions, good classification effect, accuracy be high.
Further, because convolutional neural networks local weight is shared, so utilizing convolutional neural networks in the process of element Images Classification, can reduce computation complexity, classification effectiveness is high.
See the structural representation that Fig. 5, Fig. 5 are a kind of part classification devices that fifth embodiment of the invention provides.As shown in Figure 5, a kind of part classification device 500 that fifth embodiment of the invention provides can comprise: at least one bus 501, at least one processor 502 be connected with bus and at least one storer 503 be connected with bus.
Wherein, processor 502, by bus 501, calls the code stored in storer 503 and inputs the convolutional neural networks after training for by part drawing picture to be sorted, and calculate the advanced features of described part drawing picture; The probability utilizing described advanced features to calculate described part drawing picture to belong to each classification; Get the classification that classification corresponding to probability maximum in described probability is described part drawing picture.
Wherein, convolutional neural networks is a kind of degree of depth learning network, convolutional neural networks has unique superiority with the special construction that its local weight is shared in image procossing, its layout is closer to the biological neural network of reality, weights share the complicacy reducing network, and particularly the image of multidimensional input vector directly can input the complexity that this feature of network avoids data reconstruction in feature extraction and assorting process.When utilizing convolutional neural networks to classify to part drawing picture, first need to utilize a large amount of sample training part classification devices, and then utilize the convolutional neural networks trained to classify to part drawing picture.Because convolutional neural networks can learn part drawing as feature at all levels, comprise low-level feature and advanced features, and the advanced features of image does not affect by photographed scene, even if the part drawing picture also namely taken under the scene of complexity also can utilize advanced features to identify part drawing picture, thus utilize this feature can carry out Classification and Identification to part drawing picture exactly.So, two processes can be divided into, i.e. the advanced features of computing element image during the classification of convolutional neural networks computing element image, and the classification utilizing this advanced features computing element image.
Wherein, part drawing similarly is the image at certain position of components referring to intercept from pcb board, in general, this image is the part drawing picture comprising this element, if but when element missing part, this part drawing picture does not also likely comprise element, or when component card is wrong, this part drawing picture also likely comprises the part drawing picture of other element;
The classification of part drawing picture refers to the classification annotation carried out part drawing picture according to the difference of component kind on pcb board, as pcb board having 100 kinds of elements, then the classification of part drawing picture has 100 kinds altogether at least, numeral between available 1-100 carries out class indication to it, also can carry out class indication with other symbol to different elements image category.
Wherein, the probable value that test pattern belongs to each classification illustrates the possible class that this test pattern belongs to each classification, obviously, probability is larger, it is larger that expression belongs to such other possibility, thus to get classification corresponding to probability maximum in the probability of each classification be the classification of test pattern, classification results will be made the most accurate.
Alternatively, in possible embodiments more of the present invention, described processor 502 by test pattern input through training after convolutional neural networks before, described processor 502 also for:
Template matches is utilized to obtain the position of element in described part drawing picture and align to described part drawing picture;
Described part drawing picture is normalized, performs described step part drawing picture being inputted the convolutional neural networks after training to trigger.
Alternatively, in possible embodiments more of the present invention, described processor 502 by part drawing picture input through training after convolutional neural networks before, described processor 502 also for:
Create the sample set of described part drawing picture;
Utilize convolutional neural networks described in image recognition database pre-training, obtain described convolutional neural networks initial parameter, described image recognition database comprises other natural image various types of collected from nature;
Based on described convolutional neural networks initial parameter, utilize described sample set to train described convolutional neural networks to finely tune described convolutional neural networks initial parameter further, and it is described by the step of the convolutional neural networks of part drawing picture input after training to trigger execution.
Wherein, the sample set of part drawing picture refers to the part drawing picture for training convolutional neural networks collected from pcb board, image recognition database (ImageNet) be existing collect from nature comprise other image basis database various types of, although ImageNet is not electronic component data collection, but it comprises the natural image of 1,500 ten thousand band marks more than 22000 classifications, the general image feature of each level can be learnt out for pre-training convolutional neural networks, obtain good convolutional neural networks initial parameter value.
Alternatively, in possible embodiments more of the present invention, the sample set of described part drawing picture comprises:
The training sample set of described part drawing picture and the test sample book collection of described part drawing picture.
Wherein, the training sample set of part drawing picture is at training stage training convolutional neural networks, and the test sample book collection of part drawing picture is the sample set for the classifying quality at the convolutional neural networks of training stage test after training original training to practice.
Alternatively, in possible embodiments more of the present invention, the sample set that described processor 502 creates described part drawing picture comprises:
Gather printed circuit board image;
With printed circuit board (PCB) template image for reference, described printed circuit board image intercepts part drawing picture and marks to record the classification of described part drawing picture to described part drawing picture;
The sample set of described part drawing picture is gathered from described part drawing picture after mark.
Alternatively, in possible embodiments more of the present invention, described processor 502 intercepts part drawing picture in described plated circuit, comprising:
The positional information of the element above printing plate figure picture is utilized automatically to intercept part drawing picture.
Be appreciated that when after the positional information knowing element, then can automatically intercept part drawing picture according to this positional information.
Alternatively, in possible embodiments more of the present invention, the positional information of described acquisition part drawing picture by the positional information of element recorded in board-like file, or can be obtained by the positional information manually marked.
Alternatively, in possible embodiments more of the present invention, described processor 502 pairs of part drawing pictures carry out mark and comprise:
Mark according to element class information.
Alternatively, in possible embodiments more of the present invention, before described processor 502 gathers the sample set of described part drawing picture, described processor 502 also for:
Template matches is utilized to obtain the position of element in described part drawing picture and align to described part drawing picture;
Described part drawing picture is normalized, to trigger the step of the sample set performing the described part drawing picture of described collection.
Be understandable that, the function of each functional module of the part classification device 500 of the present embodiment can according to the method specific implementation in said method embodiment, and its specific implementation process with reference to the associated description of said method embodiment, can repeat no more herein.
Can find out, in the scheme of the present embodiment, part classification device 500 by the convolutional neural networks of part drawing picture to be sorted input after training, and calculates the advanced features of described part drawing picture; Part classification device 500 recycles described advanced features and calculates the probability that described part drawing picture belongs to each classification; And get the classification that classification corresponding to probability maximum in described probability is described part drawing picture.Because convolutional neural networks can learn the advanced features of part drawing picture, when the described embodiment of the present invention utilizes convolutional neural networks to classify to part drawing picture, will the collection of part drawing picture be made not by context restrictions, good classification effect, accuracy be high.
Further, because convolutional neural networks local weight is shared, so utilizing convolutional neural networks in the process of element Images Classification, can reduce computation complexity, classification effectiveness is high.
The embodiment of the present invention also provides a kind of computer-readable storage medium, and wherein, this computer-readable storage medium can have program stored therein, and comprises the part or all of step of any part classification method recorded in said method embodiment when this program performs.
It should be noted that, for aforesaid each embodiment of the method, in order to simple description, therefore it is all expressed as a series of combination of actions, but those skilled in the art should know, the present invention is not by the restriction of described sequence of movement, because according to the present invention, some step can adopt other orders or carry out simultaneously.Secondly, those skilled in the art also should know, the embodiment described in instructions all belongs to preferred embodiment, and involved action and module might not be that the present invention is necessary.
In the above-described embodiments, the description of each embodiment is all emphasized particularly on different fields, in certain embodiment, there is no the part described in detail, can see the associated description of other embodiments.
In several embodiments that the application provides, should be understood that, disclosed device, the mode by other realizes.Such as, device embodiment described above is only schematic, the such as division of described unit, be only a kind of logic function to divide, actual can have other dividing mode when realizing, such as multiple unit or assembly can in conjunction with or another system can be integrated into, or some features can be ignored, or do not perform.Another point, shown or discussed coupling each other or direct-coupling or communication connection can be by some interfaces, and the indirect coupling of device or unit or communication connection can be electrical or other form.
The described unit illustrated as separating component or can may not be and physically separates, and the parts as unit display can be or may not be physical location, namely can be positioned at a place, or also can be distributed in multiple network element.Some or all of unit wherein can be selected according to the actual needs to realize the object of the present embodiment scheme.
In addition, each functional unit in various embodiments of the present invention can be integrated in a processing unit, also can be that the independent physics of unit exists, also can two or more unit in a unit integrated.Above-mentioned integrated unit both can adopt the form of hardware to realize, and the form of SFU software functional unit also can be adopted to realize.
If described integrated unit using the form of SFU software functional unit realize and as independently production marketing or use time, can be stored in a computer read/write memory medium.Based on such understanding, the part that technical scheme of the present invention contributes to prior art in essence in other words or all or part of of this technical scheme can embody with the form of software product, this computer software product is stored in a storage medium, comprises all or part of step of some instructions in order to make a computer equipment (can be can embedded equipment personal computer, server or the network equipment etc.) perform method described in each embodiment of the present invention.And aforesaid storage medium comprises: USB flash disk, ROM (read-only memory) (ROM, Read-OnlyMemory), random access memory (RAM, RandomAccessMemory), portable hard drive, magnetic disc or CD etc. various can be program code stored medium.
The above, above embodiment only in order to technical scheme of the present invention to be described, is not intended to limit; Although with reference to previous embodiment to invention has been detailed description, those of ordinary skill in the art is to be understood that: it still can be modified to the technical scheme described in foregoing embodiments, or carries out equivalent replacement to wherein portion of techniques feature; And these amendments or replacement, do not make the essence of appropriate technical solution depart from the scope of various embodiments of the present invention technical scheme.
Claims (10)
1. a part classification method, is characterized in that, described method comprises:
By the convolutional neural networks of part drawing picture to be sorted input after training, and calculate the advanced features of described part drawing picture;
The probability utilizing described advanced features to calculate described part drawing picture to belong to each classification;
Get the classification that classification corresponding to probability maximum in described probability is described part drawing picture.
2. method according to claim 1, is characterized in that, described by part drawing picture to be sorted input through training after convolutional neural networks before, described method also comprises:
Template matches is utilized to obtain the position of element in described part drawing picture and align to described part drawing picture;
Described part drawing picture is normalized, performs described step part drawing picture being inputted the convolutional neural networks after training to trigger.
3. method according to claim 1 and 2, is characterized in that, described by part drawing picture input through training after convolutional neural networks before, described method also comprises:
Create the sample set of described part drawing picture;
Utilize convolutional neural networks described in image recognition database pre-training, obtain described convolutional neural networks initial parameter, described image recognition database comprises other natural image various types of collected from nature;
Based on described convolutional neural networks initial parameter, utilize described sample set to train described convolutional neural networks to finely tune described convolutional neural networks initial parameter further, and it is described by the step of the convolutional neural networks of part drawing picture input after training to trigger execution.
4. method according to claim 3, is characterized in that, the sample set of the described part drawing picture of described establishment comprises:
Gather printed circuit board image;
With printed circuit board (PCB) template image for reference, described printed circuit board image intercepts part drawing picture and marks to record the classification of described part drawing picture to described part drawing picture;
The sample set of described part drawing picture is gathered from described part drawing picture after mark.
5. method according to claim 4, is characterized in that, the sample set of described part drawing picture comprises:
The training sample set of described part drawing picture and the test sample book collection of described part drawing picture.
6. a part classification device, is characterized in that, described device comprises:
First computing module, for part drawing picture to be sorted is inputted the convolutional neural networks after training, and calculates the advanced features of described part drawing picture;
Second computing module, the probability utilizing described advanced features to calculate described part drawing picture to belong to each classification;
Sort module, for getting the classification that classification corresponding to probability maximum in described probability is described part drawing picture.
7. device according to claim 6, is characterized in that, described device also comprises:
Pretreatment module, obtains the position of element in described part drawing picture for utilizing template matches and aligns to described part drawing picture;
Described part drawing picture is normalized, performs described by the step of the convolutional neural networks of the part drawing picture of classification input after training to trigger described first computing module.
8. the device according to claim 6 or 7, is characterized in that, described device also comprises:
Sample creation module, for creating the sample set of described part drawing picture;
First training module, utilizes convolutional neural networks described in image recognition database pre-training, obtains described convolutional neural networks initial parameter, and described image recognition database comprises other natural image various types of collected from nature;
Second training module, for based on described convolutional neural networks initial parameter, utilize described sample set to train described convolutional neural networks to finely tune described convolutional neural networks initial parameter further, perform described by the step of the convolutional neural networks of the part drawing picture of classification input after training to trigger described first computing module.
9. device according to claim 8, is characterized in that, described sample creation module comprises:
First collecting unit, for gathering printed circuit board image;
Interception unit, for printed circuit board (PCB) template image for reference, described printed circuit board image intercepts part drawing picture and marks to record the classification of described part drawing picture to described part drawing picture;
Second collecting unit, for gathering the sample set of described part drawing picture from described part drawing picture after mark.
10. device according to claim 9, is characterized in that, the sample set of described part drawing picture comprises:
The training sample set of described part drawing picture and the test sample book collection of described part drawing picture.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510819514.4A CN105426917A (en) | 2015-11-23 | 2015-11-23 | Component classification method and apparatus |
PCT/CN2016/096747 WO2017088537A1 (en) | 2015-11-23 | 2016-08-25 | Component classification method and apparatus |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510819514.4A CN105426917A (en) | 2015-11-23 | 2015-11-23 | Component classification method and apparatus |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105426917A true CN105426917A (en) | 2016-03-23 |
Family
ID=55505115
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510819514.4A Pending CN105426917A (en) | 2015-11-23 | 2015-11-23 | Component classification method and apparatus |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN105426917A (en) |
WO (1) | WO2017088537A1 (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106529564A (en) * | 2016-09-26 | 2017-03-22 | 浙江工业大学 | Food image automatic classification method based on convolutional neural networks |
WO2017088537A1 (en) * | 2015-11-23 | 2017-06-01 | 广州视源电子科技股份有限公司 | Component classification method and apparatus |
CN107256384A (en) * | 2017-05-22 | 2017-10-17 | 汕头大学 | A kind of card recognition and method of counting based on image and signal transacting |
CN107871100A (en) * | 2016-09-23 | 2018-04-03 | 北京眼神科技有限公司 | The training method and device of faceform, face authentication method and device |
CN107886131A (en) * | 2017-11-24 | 2018-04-06 | 佛山科学技术学院 | One kind is based on convolutional neural networks detection circuit board element polarity method and apparatus |
CN108984992A (en) * | 2018-09-25 | 2018-12-11 | 郑州云海信息技术有限公司 | A kind of circuit board designing method and device |
CN109359517A (en) * | 2018-08-31 | 2019-02-19 | 深圳市商汤科技有限公司 | Image-recognizing method and device, electronic equipment, storage medium, program product |
CN109446885A (en) * | 2018-09-07 | 2019-03-08 | 广州算易软件科技有限公司 | A kind of text based Identify chip method, system, device and storage medium |
CN109800470A (en) * | 2018-12-25 | 2019-05-24 | 山东爱普电气设备有限公司 | A kind of fixed low-voltage complete switch equipment unified bus calculation method and system |
CN111191655A (en) * | 2018-11-14 | 2020-05-22 | 佳能株式会社 | Object identification method and device |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109657374B (en) * | 2018-12-25 | 2023-10-27 | 中科曙光信息产业成都有限公司 | Modeling system and modeling method for printed circuit board |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8582807B2 (en) * | 2010-03-15 | 2013-11-12 | Nec Laboratories America, Inc. | Systems and methods for determining personal characteristics |
CN103886318A (en) * | 2014-03-31 | 2014-06-25 | 武汉天仁影像科技有限公司 | Method for extracting and analyzing nidus areas in pneumoconiosis gross imaging |
CN103927534A (en) * | 2014-04-26 | 2014-07-16 | 无锡信捷电气股份有限公司 | Sprayed character online visual detection method based on convolutional neural network |
CN104036474A (en) * | 2014-06-12 | 2014-09-10 | 厦门美图之家科技有限公司 | Automatic adjustment method for image brightness and contrast |
CN104809426A (en) * | 2014-01-27 | 2015-07-29 | 日本电气株式会社 | Convolutional neural network training method and target identification method and device |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104463241A (en) * | 2014-10-31 | 2015-03-25 | 北京理工大学 | Vehicle type recognition method in intelligent transportation monitoring system |
CN104850890B (en) * | 2015-04-14 | 2017-09-26 | 西安电子科技大学 | Instance-based learning and the convolutional neural networks parameter regulation means of Sadowsky distributions |
CN104992142B (en) * | 2015-06-03 | 2018-04-17 | 江苏大学 | A kind of pedestrian recognition method being combined based on deep learning and attribute study |
CN105513046B (en) * | 2015-11-23 | 2019-03-01 | 广州视源电子科技股份有限公司 | The polar recognition methods of electronic component and system, mask method and system |
CN105469400B (en) * | 2015-11-23 | 2019-02-26 | 广州视源电子科技股份有限公司 | The quick method and system for identifying, marking of electronic component polar orientation |
CN105426917A (en) * | 2015-11-23 | 2016-03-23 | 广州视源电子科技股份有限公司 | Component classification method and apparatus |
-
2015
- 2015-11-23 CN CN201510819514.4A patent/CN105426917A/en active Pending
-
2016
- 2016-08-25 WO PCT/CN2016/096747 patent/WO2017088537A1/en active Application Filing
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8582807B2 (en) * | 2010-03-15 | 2013-11-12 | Nec Laboratories America, Inc. | Systems and methods for determining personal characteristics |
CN104809426A (en) * | 2014-01-27 | 2015-07-29 | 日本电气株式会社 | Convolutional neural network training method and target identification method and device |
CN103886318A (en) * | 2014-03-31 | 2014-06-25 | 武汉天仁影像科技有限公司 | Method for extracting and analyzing nidus areas in pneumoconiosis gross imaging |
CN103927534A (en) * | 2014-04-26 | 2014-07-16 | 无锡信捷电气股份有限公司 | Sprayed character online visual detection method based on convolutional neural network |
CN104036474A (en) * | 2014-06-12 | 2014-09-10 | 厦门美图之家科技有限公司 | Automatic adjustment method for image brightness and contrast |
Non-Patent Citations (1)
Title |
---|
章坚: ""基于计算机视觉的电表PCB板智能识别系统的研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2017088537A1 (en) * | 2015-11-23 | 2017-06-01 | 广州视源电子科技股份有限公司 | Component classification method and apparatus |
CN107871100A (en) * | 2016-09-23 | 2018-04-03 | 北京眼神科技有限公司 | The training method and device of faceform, face authentication method and device |
CN107871100B (en) * | 2016-09-23 | 2021-07-06 | 北京眼神科技有限公司 | Training method and device of face model, and face authentication method and device |
CN106529564B (en) * | 2016-09-26 | 2019-05-31 | 浙江工业大学 | A kind of food image automatic classification method based on convolutional neural networks |
CN106529564A (en) * | 2016-09-26 | 2017-03-22 | 浙江工业大学 | Food image automatic classification method based on convolutional neural networks |
CN107256384A (en) * | 2017-05-22 | 2017-10-17 | 汕头大学 | A kind of card recognition and method of counting based on image and signal transacting |
CN107886131A (en) * | 2017-11-24 | 2018-04-06 | 佛山科学技术学院 | One kind is based on convolutional neural networks detection circuit board element polarity method and apparatus |
CN109359517A (en) * | 2018-08-31 | 2019-02-19 | 深圳市商汤科技有限公司 | Image-recognizing method and device, electronic equipment, storage medium, program product |
CN109446885A (en) * | 2018-09-07 | 2019-03-08 | 广州算易软件科技有限公司 | A kind of text based Identify chip method, system, device and storage medium |
CN109446885B (en) * | 2018-09-07 | 2022-03-15 | 广州算易软件科技有限公司 | Text-based component identification method, system, device and storage medium |
CN108984992A (en) * | 2018-09-25 | 2018-12-11 | 郑州云海信息技术有限公司 | A kind of circuit board designing method and device |
CN108984992B (en) * | 2018-09-25 | 2022-03-04 | 郑州云海信息技术有限公司 | Circuit board design method and device |
CN111191655A (en) * | 2018-11-14 | 2020-05-22 | 佳能株式会社 | Object identification method and device |
CN111191655B (en) * | 2018-11-14 | 2024-04-16 | 佳能株式会社 | Object identification method and device |
CN109800470A (en) * | 2018-12-25 | 2019-05-24 | 山东爱普电气设备有限公司 | A kind of fixed low-voltage complete switch equipment unified bus calculation method and system |
Also Published As
Publication number | Publication date |
---|---|
WO2017088537A1 (en) | 2017-06-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105426917A (en) | Component classification method and apparatus | |
CN105184778B (en) | A kind of detection method and device | |
CN109978893B (en) | Training method, device, equipment and storage medium of image semantic segmentation network | |
CN106650740B (en) | A kind of licence plate recognition method and terminal | |
CN110348294A (en) | The localization method of chart, device and computer equipment in PDF document | |
CN108197668A (en) | The method for building up and cloud system of model data collection | |
CN105989330A (en) | Picture detection method and apparatus | |
CN111046879B (en) | Certificate image classification method, device, computer equipment and readable storage medium | |
CN110443159A (en) | Digit recognition method, device, electronic equipment and storage medium | |
CN109919252A (en) | The method for generating classifier using a small number of mark images | |
CN110222582B (en) | Image processing method and camera | |
CN110110726A (en) | The recognition methods of power equipment nameplate, device, computer equipment and storage medium | |
CN106611015A (en) | Tag processing method and apparatus | |
US20230215154A1 (en) | Apparatus and method for detecting elements of an assembly | |
CN110148117A (en) | Power equipments defect recognition methods, device and storage medium based on electric power image | |
CN115062186B (en) | Video content retrieval method, device, equipment and storage medium | |
CN113159014A (en) | Objective question reading method, device, equipment and storage medium based on handwritten question numbers | |
CN112036304A (en) | Medical bill layout identification method and device and computer equipment | |
CN116543261A (en) | Model training method for image recognition, image recognition method device and medium | |
CN112288699A (en) | Method, device, equipment and medium for evaluating relative definition of image | |
CN108549857A (en) | Event detection model training method, device and event detecting method | |
CN112990282B (en) | Classification method and device for fine-granularity small sample images | |
CN116522565B (en) | BIM-based power engineering design power distribution network planning method and computer equipment | |
CN112861934A (en) | Image classification method and device of embedded terminal and embedded terminal | |
CN111680553A (en) | Pathological image identification method and system based on depth separable convolution |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160323 |
|
RJ01 | Rejection of invention patent application after publication |