CN106874954A - The method and relevant apparatus of a kind of acquisition of information - Google Patents
The method and relevant apparatus of a kind of acquisition of information Download PDFInfo
- Publication number
- CN106874954A CN106874954A CN201710090311.5A CN201710090311A CN106874954A CN 106874954 A CN106874954 A CN 106874954A CN 201710090311 A CN201710090311 A CN 201710090311A CN 106874954 A CN106874954 A CN 106874954A
- Authority
- CN
- China
- Prior art keywords
- rubbish
- image
- information
- classification
- recognized
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
- G06F18/24155—Bayesian classification
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention discloses a kind of method of acquisition of information, including:Images to be recognized is obtained, the image comprising at least one rubbish to be identified in images to be recognized;Images to be recognized is processed using preset training pattern, preset training pattern is the functional relationship model of sample image and features of classification, features of classification is used to represent the corresponding rubbish classification of sample image and positional information;Result according to preset training pattern obtains the corresponding recovery classification information of rubbish to be identified;If reclaiming classification information to indicate to include target rubbish image in images to be recognized, target position information is obtained.The present invention also provides a kind of information acquisition device.The present invention is identified using preset training pattern to the image of rubbish to be identified, and the corresponding recovery classification information of rubbish to be identified in detection image, finally according to reclaiming, classification information is next to filter out target rubbish image automatically, only need to carry out one-time detection to images to be recognized, so as to lift detection efficiency.
Description
Technical field
The present invention relates to computer disposal field, more particularly to a kind of acquisition of information method and relevant apparatus.
Background technology
With the quickening and the rapid raising of living standards of the people of the fast-developing urbanization process of Chinese society economy,
City produces and is also consequently increased rapidly with the junk produced in life process, house refuse land occupation, pollution environment
Situation and the influence to health of people are also all the more obvious.How more efficiently to reclaim and process rubbish also of interest as all circles
Focus.
Need to pay attention to the classification to rubbish during rubbish is reclaimed, that is, need according to certain rule or standard
By refuse classification storage, put on by classification and carrying of classifying, so as to be transformed into public resource.At present, to the detection master of specific refuse
The classification and Detection method based on sliding window, specific classification and Detection method is depended on to refer to Fig. 1, Fig. 1 is prior art
One embodiment schematic diagram of middle sliding window detection method, the first image to different proportion detect respectively, Ran Houji
Characteristic value is calculated with by grader, last output integrated testing result.
Although being based on the classification and Detection method of sliding window using similar Fig. 1, compared to artificially being classified more to rubbish
Beneficial to saving human cost, but, the rubbish detection based on sliding window needs to take multiple scan whole image, Ji Jiangtu
It is scanned as being divided into each yardstick, and each position is classified, like this, it usually needs the detection of plurality magnitude
Number of times, causes that during detection the plenty of time can be consumed, so as to reduce detection efficiency.
The content of the invention
The method and relevant apparatus of a kind of acquisition of information are the embodiment of the invention provides, using preset training pattern to bag
The image for having contained at least one rubbish to be identified is identified, and the corresponding recovery classification letter of rubbish to be identified in detection image
Breath, finally according to reclaiming classification information to filter out target rubbish image automatically, only needs to carry out one to images to be recognized with this
Secondary detection, is greatly improved detection speed, so as to lift detection efficiency.
In view of this, first aspect present invention provides a kind of method of acquisition of information, including:
Images to be recognized is obtained, the image comprising at least one rubbish to be identified in the images to be recognized;
The images to be recognized is processed using preset training pattern, wherein, the preset training pattern is sample
The functional relationship model of image and features of classification, the features of classification is used to represent the corresponding rubbish class of the sample image
Other and positional information;
According to the result of the preset training pattern, rubbish correspondence to be identified described in the images to be recognized is obtained
Recovery classification information;
If the recovery classification information indicates to include target rubbish image in the images to be recognized, the target is obtained
The corresponding target position information of rubbish image.
It is described to use preset training with reference to the embodiment of the present invention in a first aspect, in the first possible implementation
Before model is processed the images to be recognized, methods described also includes:
Training sample image set is treated in acquisition, wherein, it is described to treat in training sample image set comprising at least one sample
Image, the sample image includes at least one sub-district area image;
The corresponding rubbish classification and the positional information are determined according to the sub-district area image;
The rubbish classification and the positional information according to the sample image corresponding to sub-district area image
Obtain the preset training pattern.
With reference to the first possible implementation of the first aspect of the embodiment of the present invention, in second possible implementation
In, it is described that the corresponding rubbish classification and the positional information are determined according to the sub-district area image, including:
Pre-set categories information is received, the rubbish classification is included in the pre-set categories information;
The corresponding positional information of target rubbish is made in sub-district area image acceptance of the bid, the positional information includes institute
State at least one in center of the target rubbish in the sub-district area image, height, width and area coincidence factor.
The first or second of first aspect, first aspect with reference to the embodiment of the present invention may implementation, the
It is described the images to be recognized is processed using preset training pattern in three kinds of possible implementations, including:
The images to be recognized is input into the preset training pattern, wherein, the images to be recognized includes at least one
Individual sub-district area image to be identified;
Obtain the corresponding class probability value of the sub-district area image to be identified;
The result according to the preset training pattern, obtains rubbish to be identified described in the images to be recognized
Corresponding recovery classification information, including:
If the class probability value is more than or equal to pre-determined threshold, it is determined that treated described in the sub-district area image to be identified
The corresponding recovery classification information of identification rubbish.
With reference to the embodiment of the present invention in a first aspect, in the 4th kind of possible implementation, if the recovery class
Other information indicates to include target rubbish image in the images to be recognized, then obtain the corresponding target position of the target rubbish image
Confidence ceases, including:
If being determined in the images to be recognized comprising the target rubbish image according to the recovery classification information, detect
The area coincidence factor of the target rubbish image;
According to the area coincidence factor and the preset training pattern of the target rubbish image, the target is obtained
At least one in center of the rubbish image in the images to be recognized, height and width.
Second aspect present invention provides a kind of information acquisition device, including:
First acquisition module, for obtaining images to be recognized, includes at least one rubbish to be identified in the images to be recognized
The image of rubbish;
Processing module, for being entered to the images to be recognized that first acquisition module is obtained using preset training pattern
Row treatment, wherein, the preset training pattern is the functional relationship model of sample image and features of classification, the category feature
It is worth for representing the corresponding rubbish classification of the sample image and positional information;
Second acquisition module, for the result according to the preset training pattern, in the acquisition images to be recognized
The corresponding recovery classification information of the rubbish to be identified;
3rd acquisition module, if waiting to know described in the recovery classification information instruction obtained for second acquisition module
Target rubbish image is included in other image, then obtains the corresponding target position information of the target rubbish image.
With reference to the second aspect of the embodiment of the present invention, in the first possible implementation, described device also includes:
4th acquisition module, for the processing module using preset training pattern to the images to be recognized at
Before reason, training sample image set is treated in acquisition, wherein, it is described to treat in training sample image set comprising at least one sample graph
Picture, the sample image includes at least one sub-district area image;
Determining module, for determining the corresponding rubbish classification and position letter according to the sub-district area image
Breath;
5th acquisition module, for the sub-district area image institute that the determining module according to the sample image determines
The corresponding rubbish classification and the positional information obtain the preset training pattern.
With reference to the first possible implementation of the second aspect of the embodiment of the present invention, in second possible implementation
In, the determining module includes:
Receiving unit, for receiving pre-set categories information, includes the rubbish classification in the pre-set categories information;
Unit is demarcated, it is described for making the corresponding positional information of target rubbish in sub-district area image acceptance of the bid
Positional information includes center of the target rubbish in the sub-district area image, height, width and area coincidence factor
In at least one.
The first or second of second aspect, second aspect with reference to the embodiment of the present invention may implementation, the
In three kinds of possible implementations, the processing module includes:
Input block, for the images to be recognized to be input into the preset training pattern, wherein, the figure to be identified
As including at least one sub-district area image to be identified;
First acquisition unit, for obtaining the corresponding class probability value of the sub-district area image to be identified;
Second acquisition module includes:
Determining unit, if being more than or equal to pre- gating for the class probability value that the first acquisition unit is obtained
Limit, it is determined that the corresponding recovery classification information of rubbish to be identified described in the sub-district area image to be identified.
With reference to the second aspect of the embodiment of the present invention, in the 4th kind of possible implementation, the 3rd acquisition module
Including:
Detection unit, if for being determined in the images to be recognized comprising the target rubbish according to the recovery classification information
Rubbish image, then detect the area coincidence factor of the target rubbish image;
Second acquisition unit, the area of the target rubbish image for being obtained according to detection unit detection
Coincidence factor and the preset training pattern, obtain center in the images to be recognized of the target rubbish image,
Height and width at least one.
As can be seen from the above technical solutions, the embodiment of the present invention has advantages below:
In the embodiment of the present invention, there is provided a kind of method of acquisition of information, images to be recognized is obtained first, the figure to be identified
The image of at least one rubbish to be identified is contained as in, images to be recognized is processed using preset training pattern then,
Wherein, preset training pattern is the functional relationship model of sample image and features of classification, and features of classification is used to represent sample
The corresponding rubbish classification of image and positional information.Then according to the result of preset training pattern, images to be recognized is obtained
In the corresponding recovery classification information of rubbish to be identified, if reclaim classification information to indicate to contain target rubbish in images to be recognized
Image, then can just obtain the corresponding target position information of target rubbish image.Through the above way, using preset training
Model is identified to the image for containing at least one rubbish to be identified, and corresponding time of rubbish to be identified in detection image
Classification information is received, finally according to reclaiming classification information to filter out target rubbish image automatically, is only needed to figure to be identified with this
As carrying out one-time detection, detection speed is greatly improved, so as to lift detection efficiency.
Brief description of the drawings
Fig. 1 is one embodiment schematic diagram of sliding window detection method in the prior art;
Fig. 2 is a schematic flow sheet of information acquisition method in the embodiment of the present invention;
Fig. 3 is one embodiment schematic diagram of statistical-simulation spectrometry model in the embodiment of the present invention;
Fig. 4 is method one embodiment schematic diagram of acquisition of information in the embodiment of the present invention;
Fig. 5 is to obtain a network architecture diagram of preset training pattern in the embodiment of the present invention;
Fig. 6 is a schematic diagram of acquisition target rubbish image in the embodiment of the present invention;
Fig. 7 is another schematic diagram of acquisition target rubbish image in the embodiment of the present invention;
Fig. 8 is information acquisition device one embodiment schematic diagram in the embodiment of the present invention;
Fig. 9 is another embodiment schematic diagram of information acquisition device in the embodiment of the present invention;
Figure 10 is another embodiment schematic diagram of information acquisition device in the embodiment of the present invention;
Figure 11 is another embodiment schematic diagram of information acquisition device in the embodiment of the present invention;
Figure 12 is another embodiment schematic diagram of information acquisition device in the embodiment of the present invention.
Specific embodiment
The method and relevant apparatus of a kind of acquisition of information are the embodiment of the invention provides, using preset training pattern to bag
The image for having contained at least one rubbish to be identified is identified, and the corresponding recovery classification letter of rubbish to be identified in detection image
Breath, finally according to reclaim classification information come filter out target rubbish image automatically, be instead of with this manually classify it is to be identified
Rubbish, so as to significantly save human cost.
Term " first ", " second ", " the 3rd ", " in description and claims of this specification and above-mentioned accompanying drawing
Four " etc. (if present) is for distinguishing similar object, without for describing specific order or precedence.Should manage
Solution so data for using can be exchanged in the appropriate case, so that embodiments of the invention described herein for example can be removing
Order beyond those for illustrating herein or describing is implemented.Additionally, term " comprising " and " having " and theirs is any
Deformation, it is intended that covering is non-exclusive to be included, for example, containing process, method, system, the product of series of steps or unit
Product or equipment are not necessarily limited to those steps clearly listed or unit, but may include not list clearly or for this
A little processes, method, product or other intrinsic steps of equipment or unit.
It should be understood that the embodiment of the present invention is mainly used in intelligent garbage category identification system, the system can be to input
Rubbish recognized in real time, detects not recyclable rubbish, so as to carry out warning prompt to the not recyclable rubbish for putting into.
Specifically, Fig. 2 is referred to, Fig. 2 is a schematic flow sheet of information acquisition method in the embodiment of the present invention, is such as schemed
Shown, in a step 101, the camera being deployed in intelligent garbage category identification system shoots images to be recognized, then in step
Background process is carried out to images to be recognized in rapid 102, will be during images to be recognized passes through the training pattern that pre-sets, by this
Model exports corresponding result.In step 103, the result for being exported according to training pattern is detected in images to be recognized
With the presence or absence of the image of not recyclable rubbish, if it does not exist, then into step 104, conversely, then jumping to step 105.If
Image of the images to be recognized without not recyclable rubbish, then at step 104 by detection, without being moved to not recyclable rubbish
Except treatment.If however, there is the image of not recyclable rubbish in images to be recognized, basis can not be returned in step 105
Receive rubbish image to this not recyclable rubbish position, so as to reach remove the rubbish purpose.
Of the invention main using image steganalysis method, in actual applications, image steganalysis method can be included
It is not limited to statistical pattern recognition method, structure model recognition method, Fuzzy Pattern Recognition Method and neutral net mould
Formula recognition methods.Above-mentioned four kinds of recognition methods will be introduced below:
The first is statistical pattern recognition method, statistical-simulation spectrometry be it is most ripe at present be also most widely used side
Method, it mainly solves the problems, such as optimum classifier using Bayes decision rule.The basic thought of statistical decision theory is exactly not
A decision boundary is set up in same pattern class, one given pattern is included into corresponding pattern class using decision function.
Wherein, the basic model of statistical-simulation spectrometry is as shown in figure 3, Fig. 3 is of statistical-simulation spectrometry model in the embodiment of the present invention
Individual embodiment schematic diagram, the model mainly includes two kinds of operation models:Training operation model and sort operation model, wherein training
Operation model mainly has sample to complete the division to decision boundary using oneself, and takes certain study mechanism to ensure to be based on
The division of sample is optimal;And the decision-making letter that sort operation model is mainly got to the pattern being input into using its feature and training
Number and in mode division to corresponding modes class.
Statistical pattern recognition method sets up statistical-simulation spectrometry model based on decision theory mathematically.Its basic mould
Type is:A large amount of statistical analyses are carried out to studied image, the understanding of regularity is found out, and selects the feature of reflection image essence
Carry out Classification and Identification.Statistical-simulation spectrometry system can be divided into two kinds of operational modes:Training operational mode and sort run pattern.Instruction
Practice in operational mode, pretreatment module is responsible for splitting feature interested from background, removes noise and carry out it
It is operated;Characteristic selecting module is mainly responsible for finding suitable feature representing input pattern;It is special that grader is responsible for training segmentation
Levy space.In sort run pattern, input pattern is assigned to certain and referred to by the grader being trained to according to the feature of measurement
Fixed class.Statistical-simulation spectrometry composition is as shown in Figure 2.
Second is structure model recognition method, for more complicated pattern, such as using the method for statistical-simulation spectrometry, institute
The difficulty for facing is exactly the problem of feature extraction, and the characteristic quantity required by it is very huge, some complex patterns
Accurate classification is highly difficult, so as to naturally enough just expect a kind of such design, i.e., hardy a complex patterns is divided into
The combination of some simpler subpatterns, and subpattern is divided into some primitives, by the identification to primitive, and then recognizes submodule
Formula, finally recognizes the complex patterns.As english sentence is by some phrases, again by word, word constitutes one to phrase by letter again
Sample.The language of the structure of pattern, referred to as schema description language are described with one group of pattern primitive and their composition.Domination primitive
The rule of compositional model is referred to as the syntax.After each primitive is identified, whole pattern can be just made using syntactic analysis and is known
Not.Whether certain specific syntax is met with this sentence, to differentiate whether it belongs to a certain classification.Here it is configuration mode identification
Basic thought.
Configuration mode identification system is main to be made up of several parts such as pretreatment, primitive extraction, syntactic analysis and grammatical inferences.
By the pattern of pretreatment segmentation, the primitive string (i.e. character string) to form description pattern is extracted through primitive.Syntactic analysis is according to the syntax
The syntax that reasoning is inferred, adjudicate the pattern class described by orderly character string, obtain court verdict.Different pattern class correspondences
The different syntax, different targets are described.For the syntax for obtaining being adapted in pattern class, similar to statistical-simulation spectrometry
Training process, it is necessary to gather enough training mode samples in advance, extracts, corresponding grammatical inference out through primitive.It is real
Apply also certain difficulty in border.
The third is Fuzzy Pattern Recognition Method, and the theoretical foundation of Fuzzy Pattern Recognition is the mould that the sixties in 20th century is born
Paste mathematics, the thinking logic that it is recognized according to people to things with reference to the characteristics of human brain identification things, will be commonly used in computer
Two-valued function turn to continuous logic.In field of image recognition application, the method can simplify image identification system, and have
Practical and reliable the features such as.
Pattern-recognition is a frontier branch of science, and it and many technology-oriented disciplines have close contact, inherently manually
The important component of intelligence, therefore, essentially, the key problem that pattern-recognition will be discussed is how to make machine
Device can simulate the method for thinking of human brain, objective things are carried out with effective identification and classification.On the one hand it is existing widely to use
Statistical pattern recognition method compared with human brain carries out pattern-recognition, its difference is also very big, on the other hand objective thing to be identified
Thing often has different degrees of ambiguity again.
Many scholars attempt to solve pattern recognition problem with the method for fuzzy mathematics, form a particular study neck
Domain --- Fuzzy Pattern Recognition.The theoretical and method of comparative maturity has maximum to belong to principle, the pattern based on fuzzy equivalence relation
Classification, pattern classification and fuzzy clustering based on fuzzy resembling relation, wherein the research of fuzzy clustering method and application particularly into
Work(and extensively.At present, Fuzzy Pattern Recognition Method has been widely used figure identification, chromosome and white blood cell identification, image object
Shape analysis and handwritten text identification etc., but wherein also run into many difficulties, one of them typical example is exactly to be subordinate to
The determination of function is often with experience color.The key for carrying out image recognition using blur method is to determine being subordinate to for a certain classification
Function, and all kinds of statistical indicators will be then that degree of membership is total to by the value of the gray value of sample pixel and the membership function of sample pixel
With decision.Degree of membership represents that object is subordinate to the degree of a certain class.
4th kind is recognition method of neural network patterns, and the research of neutral net starts from the forties in 20th century, last century 80
Age starts to be risen extensively in various countries, and neural network filter comes from the research to animal nervous system, by using hardware
Or the method for software, many is established with a large amount of processing units as node, each unit realizes opening up for interconnection by certain pattern
Rush the net network.The network is capable of the 26S Proteasome Structure and Function of apish nervous system by certain mechanism.
Neutral net is a kind of brand-new mode identification technology, the characteristics of it has the following aspects:
(1) the characteristics of neutral net has distributed storage information;
(2) neuron can operation independent and the information that receives for the treatment of, i.e. system be capable of the information of parallel processing input;
(3) ability with self-organizing and self study.
It is understood that the embodiment of the present invention is mainly identified using the method for neural network filter to rubbish
And positioning, the neutral net is specifically as follows convolutional neural networks (Convolutional Neural Network, CNN),
Practical application exists, and rubbish can also be identified using other neutral nets, only one signal herein, and should not be understood
To be limitation of the invention.
The method of acquisition of information in the present invention will be introduced below, refer to Fig. 4, information is obtained in the embodiment of the present invention
The method one embodiment for taking includes:
201st, images to be recognized is obtained, the image comprising at least one rubbish to be identified in images to be recognized;
In the present embodiment, information acquisition device includes at least one camera, and the camera is used for shooting at least one to be treated
Identification image.Wherein, the image of at least one rubbish to be identified, pending dustbin image can be included in images to be recognized
At least piece image shot to same rubbish to be identified can be included, for example, photographed the front view of rubbish to be identified, bowed
View and left view.
202nd, images to be recognized is processed using preset training pattern, wherein, preset training pattern is sample image
With the functional relationship model of features of classification, features of classification is for representing the corresponding rubbish classification of sample image and position letter
Breath;
In the present embodiment, information acquisition device is processed the images to be recognized for photographing using preset training pattern,
Will images to be recognized be input into preset training pattern, be then calculated corresponding result by model.
Wherein, the main functional relation for including sample image and features of classification, each sample in preset training pattern
Image is all pre-entered, the features of classification for obtaining each sample image by deep learning.Features of classification can
To represent rubbish classification and the positional information corresponding to sample image, for example, rubbish classification can be divided into three kinds, respectively
Rubbish classification A, rubbish classification B and rubbish classification C, then to the rubbish in the presence of these three rubbish classifications in each sample image
Location position is carried out, so as to obtain the positional information of rubbish in sample image.
203rd, according to the result of preset training pattern, the corresponding recovery class of rubbish to be identified in images to be recognized is obtained
Other information;
In the present embodiment, according to the result of preset training pattern, also can just obtain corresponding to the images to be recognized
Reclaim classification information.Wherein, reclaim classification information has incidence relation with rubbish classification set in advance, such as, and rubbish classification
Rubbish classification A, rubbish classification B and rubbish classification C can be divided into, and only rubbish classification A belongs to recyclable rubbish, rubbish classification
B and rubbish classification C belong to not recyclable rubbish, and information acquisition device has rubbish classification B in images to be recognized is detected
When, it is believed that reclaim classification information and belong to " not recyclable class ", by that analogy, there is rubbish in images to be recognized is detected
During classification C, it is believed that reclaim classification information and belong to " not recyclable class ", there is rubbish classification in images to be recognized is detected
During A, it is believed that reclaim classification information and belong to " recyclable class ".
It should be noted that in actual applications, rubbish classification can also have other mode classifications, such as according to material
Classified, or classified according to composition, or classified according to chemical property, do not limited herein.
If the 204, reclaiming classification information to indicate to include target rubbish image in images to be recognized, target rubbish image is obtained
Corresponding target position information.
In the present embodiment, if according to classification information is reclaimed, determining to include the mesh for needing to be extracted in images to be recognized
Mark rubbish image, then can further go to obtain the corresponding physical location of target rubbish image, i.e. target position information.
For example, information acquisition device has detected rubbish X and rubbish Y in images to be recognized, wherein, rubbish X belongs to rubbish
Rubbish classification A, corresponding recovery classification information is " recyclable class ", and rubbish Y belongs to rubbish classification B, corresponding recovery classification letter
Cease is " not recyclable class ", it is assumed that we need to choose from recyclable rubbish to be mingled in not recyclable rubbish therein, then
Image of the classification information for the rubbish Y of " not recyclable class " can will be reclaimed as target rubbish image, so as to further obtain
Position of the target rubbish image in images to be recognized, is recently determined out actual target position information.
Wherein, target position information can be represented using longitude and latitude, or be represented using coordinate system, or combine this two
The method of kind is represented, not limited herein.
In the embodiment of the present invention, there is provided a kind of method of acquisition of information, images to be recognized is obtained first, the figure to be identified
The image of at least one rubbish to be identified is contained as in, images to be recognized is processed using preset training pattern then,
Wherein, preset training pattern is the functional relationship model of sample image and features of classification, and features of classification is used to represent sample
The corresponding rubbish classification of image and positional information.Then according to the result of preset training pattern, images to be recognized is obtained
In the corresponding recovery classification information of rubbish to be identified, if reclaim classification information to indicate to contain target rubbish in images to be recognized
Image, then can just obtain the corresponding target position information of target rubbish image.Through the above way, using preset training
Model is identified to the image for containing at least one rubbish to be identified, and corresponding time of rubbish to be identified in detection image
Classification information is received, finally according to reclaiming classification information to filter out target rubbish image automatically, is only needed to figure to be identified with this
As carrying out one-time detection, detection speed is greatly improved, so as to lift detection efficiency.
Alternatively, on the basis of the corresponding embodiments of above-mentioned Fig. 4, the method for acquisition of information provided in an embodiment of the present invention
In first alternative embodiment, before being processed images to be recognized using preset training pattern, can also include:
Training sample image set is treated in acquisition, wherein, treat in training sample image set comprising at least one sample image,
Sample image includes at least one sub-district area image;
Corresponding rubbish classification and positional information are determined according to sub-district area image;
Rubbish classification and positional information according to corresponding to sample image neutron area image obtain preset training pattern.
In the present embodiment, the training of recommended information acquisition device is obtained the mode of preset training pattern.
Specifically, information acquisition device needs acquisition to treat training sample image collection comprising at least one sample image first
Close, it is generally the case that in order to the accuracy of training is needed using more sample image, and sample image is evenly divided for
It is at least one sub-district area image, carrying out learning training for every sub-regions image can obtain more accurately preset training mould
Type.This is because the small image in region is easier to catch characteristic point, so as to be conducive to the reliability of training.Most use deep learning
Method can just obtain each corresponding rubbish classification of sample image neutron area image and positional information.
Wherein, deep learning is a branch of machine learning, and it is used comprising labyrinth or by multiple non-linear
Converting the multiple process layers for constituting carries out the algorithm of higher level of abstraction to data.Deep learning is also a kind of based on right in machine learning
The method that data carry out representative learning.Observation (such as one sample image) can be represented using various ways, such as each
A series of vector of pixel intensity value, or more abstractively represent sides, or a series of given shapes region etc., do not do herein
Limit.And be easier to be learnt from example using some specific identification methods, for example, the identification of rubbish positional information or
Identification of person's rubbish classification etc..
Finally, the corresponding rubbish classification of each sample image and positional information are being got using the method for deep learning
Afterwards, it is possible to set up preset training pattern, contained between sample image and features of classification in the preset training pattern
Corresponding relation.
Additionally, be using a benefit of deep learning, can with the feature learning of non-supervisory formula or Semi-supervised and
Layered characteristic extracts highly effective algorithm to replace manual extraction feature, so that lifting feature extraction efficiency, also just can be more efficiently
Get the features of classification corresponding to different sample images.
Secondly, in the embodiment of the present invention, information acquisition device needs in advance to be trained preset training pattern, i.e., first obtain
Take and treat training sample image set, wherein, this is treated in training sample image set comprising at least one sample image, and sample graph
As including at least one sub-district area image, corresponding rubbish classification and positional information are then determined according to sub-district area image,
Preset training pattern is obtained finally according to the rubbish classification corresponding to sub-district area image and positional information.Through the above way,
More accurately preset training pattern can be learnt to obtain, be that the later use preset training pattern is entered to pending dustbin image
Row treatment provides reliable foundation, at the same time, by the way of great amount of samples image is learnt and trained, can keep away
Exempt from cumbersome engineer, and be conducive to improving the precision classified.
Alternatively, on the basis of the corresponding one embodiment of above-mentioned Fig. 4, acquisition of information provided in an embodiment of the present invention
Second alternative embodiment of method in, corresponding rubbish classification and positional information are determined according to sub-district area image, can wrap
Include:
Pre-set categories information is received, rubbish classification is included in pre-set categories information;
The corresponding positional information of target rubbish is made in the acceptance of the bid of sub-district area image, positional information is comprising target rubbish in sub-district
At least one in center, height, width and area coincidence factor in area image.
In the present embodiment, information acquisition device can in advance receive the pre-set categories information of user input, pre-set categories letter
Rubbish classification is included in breath.It should be noted that the number of rubbish classification can be three kinds or five kinds, in actual applications,
The rubbish classification of other quantity can also be set, the number not to rubbish classification is defined herein.Wherein, rubbish classification can be with
It is divided into Harmful Waste, organic waste and inorganic refuse etc., Harmful Waste refer mainly to refuse battery, paint, fluorescent tube and Expired drug etc.,
These solid waste can cause real or potentially hazardous to health or natural environment.Organic waste is in natural conditions
Under labile rubbish, it is high and perishable by moisture content mainly by the food residue of the generations such as family, restaurant and unit dining room
The rubbish of the labile organic compound composition for losing.It is recyclable that inorganic refuse refers mainly to waste paper, cullet, waste plastics and old metal etc.
The goods and materials for utilizing.
Then by sub-district area image get the bid make the corresponding positional information of target rubbish, it is possible to use rubbish classification with
And positional information obtains preset training pattern, center of the positional information comprising target rubbish in sub-district area image, height,
At least one in width and area coincidence factor.It is described in detail below and how determines position using the method for deep learning
Information.
First by large-scale public image database training CNN, with back-propagation algorithm undated parameter p20As sample graph
As the initial value of training.Then it is d × d sub-regions images the sample image containing specific refuse to be evenly dividing, and sets C as special
The number of different rubbish classification, while marking out the positional information of target rubbish in training sample image, target rubbish is specially
Specific refuse, such as bottle, display or box etc., do not limit herein.The positional information of target rubbish be expressed as (x, y,
W, H, f), rubbish classification is (p1,p2,…pC), wherein, x represents abscissa positions of the target rubbish in sub-district area image, y tables
Show ordinate position of the target rubbish in sub-district area image, W represents width of the target rubbish in sub-district area image, and H is represented
Height of the target rubbish in sub-district area image, f represents the prediction area of target rubbish and the area coincidence factor of real area.
(x, y) then represents target rubbish in the normalized center in current sub-region, 0 < x < 1, and 0 < y < 1.F is then
It is embodied asA represents target rubbish prediction area, and B represents the real area of target rubbish.(W, H) represents mesh
Mark the wide and height of the relatively whole subregion image normalization of rubbish, 0 < W < 1, and 0 < H < 1.
Finally, output layer is replaced by d × d × (5+C) multidimensional tensor output, is calculated by backpropagation by new training data
Method updates CNN parameters, will parameter p20It is updated to p21。
Specifically, Fig. 5 is referred to, Fig. 5 is the network architecture diagram that preset training pattern is obtained in the embodiment of the present invention,
Wherein, the network architecture is specially CNN, and CNN is made up of the full articulamentum on one or more convolutional layers and top, while also wrapping
Associated weights layer and pond layer are included, this structure enables that CNN is trained using the two-dimensional structure of input data.
CNN frameworks can be designed first, i.e., including the convolution number of plies, receptive field size, convolution unit number, maximum pond
Layer, convolutional layer receptive field size and pond layer step-length etc., network-related parameters refer to table 1.
Table 1
Wherein, all of target rubbish is detected using single network fl transmission.Specifically, we design CNN framves first
Structure (total number of plies, the convolution number of plies, the change number of plies, convolutional layer receptive field size, pond layer step-length etc.), the network architecture is referring to Fig. 5, network
Parameter is referring to table 1.Wherein each convolutional layer is followed by activation primitive (Rectified Linear Units, Relu) nonlinear activation
Function:F (x)=max (0, x).
Again, in the embodiment of the present invention, information acquisition device can receive pre-set categories information, be wrapped in pre-set categories information
Classification containing rubbish, and the corresponding positional information of target rubbish is made in the acceptance of the bid of sub-district area image, positional information includes target rubbish
At least one in center of the rubbish in sub-district area image, height, width and area coincidence factor, thereby determines that out sub-district
The corresponding rubbish classification of area image and positional information.Through the above way, more accurate parameter, i.e. rubbish class can be obtained
Other and positional information, recycles CNN to be trained rubbish classification and positional information, compared with other deep learning structures, energy
Access more accurately training result, that is to say, that compared to other depth algorithms, feedforward neural network and convolutional Neural net
Network needs the parameter estimated less, so as to improve the practicality and operability of scheme.
Alternatively, on the basis of above-mentioned Fig. 4 and Fig. 4 corresponding first or second embodiment, the present invention is implemented
Example provide acquisition of information the 3rd alternative embodiment of method in, using preset training pattern to images to be recognized at
Reason, can include:
Images to be recognized is input into preset training pattern, wherein, images to be recognized includes at least one sub-district to be identified
Area image;
Obtain the corresponding class probability value of sub-district area image to be identified;
According to the result of preset training pattern, the corresponding classification that reclaims of rubbish to be identified is believed in obtaining images to be recognized
Breath, can include:
If class probability value is more than or equal to pre-determined threshold, it is determined that rubbish correspondence to be identified in sub-district area image to be identified
Recovery classification information.
In the present embodiment, images to be recognized is detected using preset training pattern specifically be may include steps of:
The first step, images to be recognized is input into preset training pattern, then exports multidimensional tensor information, example by CNN
Such as export α × α × (5+C), will sample image to be evenly dividing be α × α sub-district area images to be identified, C is rubbish classification
Number.
Second step, obtains the corresponding class probability value of sub-district area image to be identified, wherein, class probability value is intended to indicate that
Occurs the probability of rubbish to be identified in sub-district area image to be identified, for example rubbish to be identified is accounted in sub-district area image to be identified
According to 25% picture of whole image, then class probability value is 25%.
3rd step, judges whether the corresponding class probability value of sub-district area image to be identified is more than or equal to pre-determined threshold, such as
The other probable value of fruit is more than or equal to pre-determined threshold, i.e. maxc∈CP (c) >=T, it is determined that included in sub-district area image to be identified and treated
Identification rubbish, wherein, T represents pre-determined threshold, and c represents certain rubbish classification;
4th step, obtains the corresponding recovery classification information of rubbish to be identified.
Further, in the embodiment of the present invention, images to be recognized is entered using preset training pattern in information acquisition device
, it is necessary to first images to be recognized is input into preset training pattern during row treatment, wherein, images to be recognized includes at least one
Individual sub-district area image to be identified, then obtains the corresponding class probability value of sub-district area image to be identified, if class probability value is big
In or equal to pre-determined threshold, then the recovery classification information of rubbish to be identified in sub-district area image to be identified can be determined.By upper
Mode is stated, information acquisition device can utilize class probability value to obtain the recovery classification information of rubbish to be identified, thus exclude not
Meet the rubbish to be identified of condition, and the recovery classification information of these rubbish need not be obtained, so as to improve infomation detection
Efficiency and the operability of detection.
Alternatively, on the basis of corresponding 3rd embodiment of above-mentioned Fig. 4, acquisition of information provided in an embodiment of the present invention
The 4th alternative embodiment of method in, if reclaim classification information indicate images to be recognized in include target rubbish image, obtain
The corresponding target position information of target rubbish image is taken, can be included:
If determining, comprising target rubbish image in images to be recognized, to detect target rubbish image according to classification information is reclaimed
Area coincidence factor;
According to the area coincidence factor and preset training pattern of target rubbish image, target rubbish image is obtained to be identified
At least one in center, height and width in image.
In the present embodiment, after the corresponding recovery classification information of rubbish to be identified is got, can be according to recovery classification
Information determines whether the rubbish to be identified belongs to target rubbish, if it is, further obtaining target rubbish image to be identified
Area coincidence factor in image, target rubbish image is positioned in images to be recognized using preset training pattern and area coincidence factor
In position, the position include center, height and width at least one.
Wherein it is possible to the repeat region that same target rubbish is detected is merged using non-maxima suppression, according to (x, y, W,
H), x represents abscissa positions of the target rubbish in images to be recognized, and y represents vertical seat of the target rubbish in images to be recognized
Cursor position, W represents width of the target rubbish in images to be recognized, and H represents height of the target rubbish in images to be recognized.
In order to make it easy to understand, referring to Fig. 6, Fig. 6 is a signal of acquisition target rubbish image in the embodiment of the present invention
Figure, it is assumed that target rubbish is bottle, and first Fig. 6 can be input into preset training pattern, is then divided into several to wait to know Fig. 6
Small pin for the case area image, then obtain the corresponding class probability value of each sub-district area image to be identified respectively, be more than in class probability value or
The corresponding recovery classification information of rubbish to be identified is obtained in the case of equal to pre-determined threshold.Because we need to find bottle, because
This bottle is target rubbish, searches which rubbish to be identified belongs to thus according to the corresponding classification information that reclaims of rubbish to be identified
Target rubbish, and then obtain the corresponding target rubbish image of target rubbish.Last preset training pattern is according to target rubbish image
Its positional information is determined, to extract the target rubbish.
Fig. 7 is another schematic diagram of acquisition target rubbish image in the embodiment of the present invention, and the specification of images to be recognized is not
Be only limitted to Fig. 6 or Fig. 7, Fig. 7 be the mode for extracting display, in actual applications, target rubbish except can be bottle or
Display, can also be other kinds of rubbish, and only one signal, is not construed as limitation of the invention herein.
Further, in the embodiment of the present invention, if determined in images to be recognized comprising mesh according to classification information is reclaimed
Mark rubbish image, then can detect the area coincidence factor of target rubbish image, the area further according to target rubbish image overlaps
Rate and preset training pattern, in determining center of the target rubbish image in images to be recognized, height and width
At least one.Through the above way, the target position information of target rubbish image can be more accurately obtained, so as to be conducive to more
The rubbish is more accurately extracted soon, thus the accuracy of detection of enhanced scheme.
The information acquisition device in the present invention is described in detail below, refers to Fig. 8, the letter in the embodiment of the present invention
Breath acquisition device 30 includes:
First acquisition module 301, it is to be identified comprising at least one in the images to be recognized for obtaining images to be recognized
The image of rubbish;
Processing module 302, it is described to be identified for what is obtained to first acquisition module 301 using preset training pattern
Image is processed, wherein, the preset training pattern is the functional relationship model of sample image and features of classification, the class
Other characteristic value is used to represent the corresponding rubbish classification of the sample image and positional information;
Second acquisition module 303, for the result according to the preset training pattern, obtains the images to be recognized
Described in the corresponding recovery classification information of rubbish to be identified;
3rd acquisition module 304, if indicating institute for the recovery classification information that second acquisition module 303 is obtained
State in images to be recognized comprising target rubbish image, then obtain the corresponding target position information of the target rubbish image.
In the present embodiment, the first acquisition module 301 obtains images to be recognized, and at least one is included in the images to be recognized
The image of rubbish to be identified, processing module 302 is using preset training pattern to being treated described in first acquisition module 301 acquisition
Identification image is processed, wherein, the preset training pattern is the functional relationship model of sample image and features of classification, institute
Features of classification is stated for representing the corresponding rubbish classification of the sample image and positional information, the second acquisition module 303
According to the result of the preset training pattern, the corresponding recovery classification of rubbish to be identified described in the images to be recognized is obtained
Information;3rd acquisition module 304, if being indicated for the recovery classification information that second acquisition module 303 is obtained described
Target rubbish image is included in images to be recognized, then obtains the corresponding target position information of the target rubbish image.
In the embodiment of the present invention, there is provided a kind of information acquisition device, the device obtains images to be recognized first, and this waits to know
The image of at least one rubbish to be identified is contained in other image, then using preset training pattern to images to be recognized at
Reason, wherein, preset training pattern is the functional relationship model of sample image and features of classification, and features of classification is used to represent sample
The corresponding rubbish classification of this image and positional information.Then according to the result of preset training pattern, figure to be identified is obtained
The corresponding recovery classification information of rubbish to be identified as in, if reclaim classification information to indicate to contain target rubbish in images to be recognized
Rubbish image, then can just obtain the corresponding target position information of target rubbish image.Through the above way, using preset instruction
Practice model to be identified the image for containing at least one rubbish to be identified, and rubbish to be identified in detection image is corresponding
Classification information is reclaimed, finally according to reclaiming classification information to filter out target rubbish image automatically, is only needed to to be identified with this
Image carries out one-time detection, and detection speed is greatly improved, so as to lift detection efficiency.
Alternatively, on the basis of the embodiment corresponding to above-mentioned Fig. 8, Fig. 9, letter provided in an embodiment of the present invention are referred to
Cease in another embodiment of acquisition device 30, described information acquisition device 30 also includes:
4th acquisition module 305, for using preset training pattern to the images to be recognized in the processing module 302
Before being processed, training sample image set is treated in acquisition, wherein, it is described to treat in training sample image set comprising at least one
Sample image, the sample image includes at least one sub-district area image;
Determining module 306, for determining the corresponding rubbish classification and the position according to the sub-district area image
Information;
5th acquisition module 307, for the subregion that the determining module 306 according to the sample image determines
The rubbish classification and the positional information corresponding to image obtain the preset training pattern.
Secondly, in the embodiment of the present invention, information acquisition device needs in advance to be trained preset training pattern, i.e., first obtain
Take and treat training sample image set, wherein, this is treated in training sample image set comprising at least one sample image, and sample graph
As including at least one sub-district area image, corresponding rubbish classification and positional information are then determined according to sub-district area image,
Preset training pattern is obtained finally according to the rubbish classification corresponding to sub-district area image and positional information.Through the above way,
More accurately preset training pattern can be learnt to obtain, be that the later use preset training pattern is entered to pending dustbin image
Row treatment provides reliable foundation, at the same time, by the way of great amount of samples image is learnt and trained, can keep away
Exempt from cumbersome engineer, and be conducive to improving the precision classified.
Alternatively, on the basis of the embodiment corresponding to above-mentioned Fig. 9, Figure 10 is referred to, it is provided in an embodiment of the present invention
In another embodiment of information acquisition device 30,
The determining module 306 includes:
Receiving unit 3061, for receiving pre-set categories information, includes the rubbish classification in the pre-set categories information;
Unit 3062 is demarcated, for making the corresponding positional information of target rubbish in sub-district area image acceptance of the bid,
The positional information includes center of the target rubbish in the sub-district area image, height, width and area weight
At least one in conjunction rate.
Again, in the embodiment of the present invention, information acquisition device can receive pre-set categories information, be wrapped in pre-set categories information
Classification containing rubbish, and the corresponding positional information of target rubbish is made in the acceptance of the bid of sub-district area image, positional information includes target rubbish
At least one in center of the rubbish in sub-district area image, height, width and area coincidence factor, thereby determines that out sub-district
The corresponding rubbish classification of area image and positional information.Through the above way, more accurate parameter, i.e. rubbish class can be obtained
Other and positional information, recycles CNN to be trained rubbish classification and positional information, compared with other deep learning structures, energy
Access more accurately training result, that is to say, that compared to other depth algorithms, feedforward neural network and convolutional Neural net
Network needs the parameter estimated less, so as to improve the practicality and operability of scheme.
Alternatively, on the basis of the embodiment corresponding to above-mentioned Fig. 8, Fig. 9 or Figure 10, Figure 11 is referred to, the present invention is real
In another embodiment of the information acquisition device 30 that example offer is provided,
The processing module 302 includes:
Input block 3021, for the images to be recognized to be input into the preset training pattern, wherein, it is described to wait to know
Other image includes at least one sub-district area image to be identified;
First acquisition unit 3022, for obtaining the corresponding class probability value of the sub-district area image to be identified;
Second acquisition module 303 includes:
Determining unit 3031, if being more than or equal to for the class probability value that the first acquisition unit is obtained default
Thresholding, it is determined that the corresponding recovery classification information of rubbish to be identified described in the sub-district area image to be identified.
Further, in the embodiment of the present invention, images to be recognized is entered using preset training pattern in information acquisition device
, it is necessary to first images to be recognized is input into preset training pattern during row treatment, wherein, images to be recognized includes at least one
Individual sub-district area image to be identified, then obtains the corresponding class probability value of sub-district area image to be identified, if class probability value is big
In or equal to pre-determined threshold, then the recovery classification information of rubbish to be identified in sub-district area image to be identified can be determined.By upper
Mode is stated, information acquisition device can utilize class probability value to obtain the recovery classification information of rubbish to be identified, thus exclude not
Meet the rubbish to be identified of condition, and the recovery classification information of these rubbish need not be obtained, so as to improve infomation detection
Efficiency and the operability of detection.
Alternatively, on the basis of the embodiment corresponding to above-mentioned Figure 11, Figure 12 is referred to, it is provided in an embodiment of the present invention
In another embodiment of information acquisition device 30,
3rd acquisition module 304 includes:
Detection unit 3041, if for being determined in the images to be recognized comprising the mesh according to the recovery classification information
Mark rubbish image, then detect the area coincidence factor of the target rubbish image;
Second acquisition unit 3042, for detecting the target rubbish image for obtaining according to the detection unit 3041
The area coincidence factor and the preset training pattern, in obtaining the target rubbish image in the images to be recognized
At least one in heart position, height and width.
Further, in the embodiment of the present invention, if determined in images to be recognized comprising mesh according to classification information is reclaimed
Mark rubbish image, then can detect the area coincidence factor of target rubbish image, the area further according to target rubbish image overlaps
Rate and preset training pattern, in determining center of the target rubbish image in images to be recognized, height and width
At least one.Through the above way, the target position information of target rubbish image can be more accurately obtained, so as to be conducive to more
The rubbish is more accurately extracted soon, thus the accuracy of detection of enhanced scheme.
It is apparent to those skilled in the art that, for convenience and simplicity of description, the system of foregoing description,
The specific work process of device and unit, may be referred to the corresponding process in preceding method embodiment, will not be repeated here.
In several embodiments provided herein, it should be understood that disclosed system, apparatus and method can be with
Realize by another way.For example, device embodiment described above is only schematical, for example, the unit
Divide, only a kind of division of logic function there can be other dividing mode when actually realizing, for example multiple units or component
Can combine or be desirably integrated into another system, or some features can be ignored, or do not perform.It is another, it is shown or
The coupling each other for discussing or direct-coupling or communication connection can be the indirect couplings of device or unit by some interfaces
Close or communicate to connect, can be electrical, mechanical or other forms.
The unit that is illustrated as separating component can be or may not be it is physically separate, it is aobvious as unit
The part for showing can be or may not be physical location, you can with positioned at a place, or can also be distributed to multiple
On NE.Some or all of unit therein can be according to the actual needs selected to realize the mesh of this embodiment scheme
's.
In addition, during each functional unit in each embodiment of the invention can be integrated in a processing unit, it is also possible to
It is that unit is individually physically present, it is also possible to which two or more units are integrated in a unit.Above-mentioned integrated list
Unit can both be realized in the form of hardware, it would however also be possible to employ the form of SFU software functional unit is realized.
If the integrated unit is to realize in the form of SFU software functional unit and as independent production marketing or use
When, can store in a computer read/write memory medium.Based on such understanding, technical scheme is substantially
The part for being contributed to prior art in other words or all or part of the technical scheme can be in the form of software products
Embody, the computer software product is stored in a storage medium, including some instructions are used to so that a computer
Equipment (can be personal computer, server, or network equipment etc.) performs the complete of each embodiment methods described of the invention
Portion or part steps.And foregoing storage medium includes:USB flash disk, mobile hard disk, read-only storage (Read-Only Memory,
ROM), random access memory (Random Access Memory, RAM), magnetic disc or CD etc. are various can be with storage program
The medium of code.
The above, the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations;Although with reference to preceding
Embodiment is stated to be described in detail the present invention, it will be understood by those within the art that:It still can be to preceding
State the technical scheme described in each embodiment to modify, or equivalent is carried out to which part technical characteristic;And these
Modification is replaced, and does not make the spirit and scope of the essence disengaging various embodiments of the present invention technical scheme of appropriate technical solution.
Claims (10)
1. a kind of method of acquisition of information, it is characterised in that including:
Images to be recognized is obtained, the image comprising at least one rubbish to be identified in the images to be recognized;
The images to be recognized is processed using preset training pattern, wherein, the preset training pattern is sample image
With the functional relationship model of features of classification, the features of classification be used for represent the corresponding rubbish classification of the sample image with
And positional information;
According to the result of the preset training pattern, corresponding time of rubbish to be identified described in the images to be recognized is obtained
Receive classification information;
If the recovery classification information indicates to include target rubbish image in the images to be recognized, the target rubbish is obtained
The corresponding target position information of image.
2. method according to claim 1, it is characterised in that it is described using preset training pattern to the images to be recognized
Before being processed, methods described also includes:
Training sample image set is treated in acquisition, wherein, it is described to treat in training sample image set comprising at least one sample image,
The sample image includes at least one sub-district area image;
The corresponding rubbish classification and the positional information are determined according to the sub-district area image;
The rubbish classification and the positional information according to the sample image corresponding to sub-district area image are obtained
The preset training pattern.
3. method according to claim 2, it is characterised in that it is described according to the sub-district area image determine it is corresponding described in
Rubbish classification and the positional information, including:
Pre-set categories information is received, the rubbish classification is included in the pre-set categories information;
The corresponding positional information of target rubbish is made in sub-district area image acceptance of the bid, the positional information includes the mesh
At least one in center of the mark rubbish in the sub-district area image, height, width and area coincidence factor.
4. according to the method in any one of claims 1 to 3, it is characterised in that described to use preset training pattern to institute
Images to be recognized is stated to be processed, including:
The images to be recognized is input into the preset training pattern, wherein, the images to be recognized is treated including at least one
Identification sub-district area image;
Obtain the corresponding class probability value of the sub-district area image to be identified;
The result according to the preset training pattern, obtains rubbish correspondence to be identified described in the images to be recognized
Recovery classification information, including:
If the class probability value is more than or equal to pre-determined threshold, it is determined that to be identified described in the sub-district area image to be identified
The corresponding recovery classification information of rubbish.
5. method according to claim 4, it is characterised in that if the recovery classification information indicate it is described to be identified
Target rubbish image is included in image, then obtains the corresponding target position information of the target rubbish image, including:
If determining that, comprising the target rubbish image in the images to be recognized, detection is described according to the recovery classification information
The area coincidence factor of target rubbish image;
According to the area coincidence factor and the preset training pattern of the target rubbish image, the target rubbish is obtained
At least one in center of the image in the images to be recognized, height and width.
6. a kind of information acquisition device, it is characterised in that including:
First acquisition module, for obtaining images to be recognized, includes at least one rubbish to be identified in the images to be recognized
Image;
Processing module, at the images to be recognized obtained to first acquisition module for the preset training pattern of use
Reason, wherein, the preset training pattern is the functional relationship model of sample image and features of classification, and the features of classification is used
In the corresponding rubbish classification of the expression sample image and positional information;
Second acquisition module, for the result according to the preset training pattern, obtains described in the images to be recognized
The corresponding recovery classification information of rubbish to be identified;
3rd acquisition module, if indicating the figure to be identified for the recovery classification information that second acquisition module is obtained
Target rubbish image is included as in, then obtains the corresponding target position information of the target rubbish image.
7. information acquisition device according to claim 6, it is characterised in that described device also includes:
4th acquisition module, for carrying out treatment to the images to be recognized using preset training pattern in the processing module
Before, training sample image set is treated in acquisition, wherein, it is described to treat in training sample image set comprising at least one sample image,
The sample image includes at least one sub-district area image;
Determining module, for determining the corresponding rubbish classification and the positional information according to the sub-district area image;
5th acquisition module, for corresponding to the sub-district area image of determining module determination according to the sample image
The rubbish classification and the positional information obtain the preset training pattern.
8. information acquisition device according to claim 7, it is characterised in that the determining module includes:
Receiving unit, for receiving pre-set categories information, includes the rubbish classification in the pre-set categories information;
Unit is demarcated, for making the corresponding positional information of target rubbish, the position in sub-district area image acceptance of the bid
Packet is containing in center of the target rubbish in the sub-district area image, height, width and area coincidence factor
At least one.
9. the information acquisition device according to any one of claim 6 to 8, it is characterised in that the processing module includes:
Input block, for the images to be recognized to be input into the preset training pattern, wherein, the images to be recognized bag
Include at least one sub-district area image to be identified;
First acquisition unit, for obtaining the corresponding class probability value of the sub-district area image to be identified;
Second acquisition module includes:
Determining unit, if being more than or equal to pre-determined threshold for the class probability value that the first acquisition unit is obtained,
Determine the corresponding recovery classification information of rubbish to be identified described in the sub-district area image to be identified.
10. information acquisition device according to claim 9, it is characterised in that the 3rd acquisition module includes:
Detection unit, if for being determined in the images to be recognized comprising the target rubbish figure according to the recovery classification information
Picture, then detect the area coincidence factor of the target rubbish image;
Second acquisition unit, the area of the target rubbish image for being obtained according to detection unit detection overlaps
Rate and the preset training pattern, obtain center, height of the target rubbish image in the images to be recognized
And at least one in width.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710090311.5A CN106874954A (en) | 2017-02-20 | 2017-02-20 | The method and relevant apparatus of a kind of acquisition of information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710090311.5A CN106874954A (en) | 2017-02-20 | 2017-02-20 | The method and relevant apparatus of a kind of acquisition of information |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106874954A true CN106874954A (en) | 2017-06-20 |
Family
ID=59166513
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710090311.5A Pending CN106874954A (en) | 2017-02-20 | 2017-02-20 | The method and relevant apparatus of a kind of acquisition of information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106874954A (en) |
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107767548A (en) * | 2017-08-01 | 2018-03-06 | 山东中公网络科技有限公司 | A kind of intelligent article retracting device, intelligent article recovery system and recovery method |
CN107967460A (en) * | 2017-12-08 | 2018-04-27 | 重庆广睿达科技有限公司 | A kind of rubbish thing based on deep neural network burns recognition methods and system |
CN108182455A (en) * | 2018-01-18 | 2018-06-19 | 齐鲁工业大学 | A kind of method, apparatus and intelligent garbage bin of the classification of rubbish image intelligent |
CN108288077A (en) * | 2018-04-17 | 2018-07-17 | 天津和或节能科技有限公司 | Grading of old paper device establishes device and method, grading of old paper system and method |
CN108665616A (en) * | 2018-05-15 | 2018-10-16 | 刘晓东 | A kind of recovery method and retracting device of rubbish |
CN108960347A (en) * | 2018-03-28 | 2018-12-07 | 中国科学院软件研究所 | A kind of recruitment evaluation system and method for convolutional neural networks handwriting recongnition Ranking Stability |
CN109308479A (en) * | 2018-09-20 | 2019-02-05 | 云南师范大学 | A kind of automatic classified reclaiming method of Campus Garbage based on deep learning |
CN109389161A (en) * | 2018-09-28 | 2019-02-26 | 广州大学 | Rubbish identification evolutionary learning method, apparatus, system and medium based on deep learning |
CN110008792A (en) * | 2018-01-05 | 2019-07-12 | 比亚迪股份有限公司 | Image detecting method, device, computer equipment and storage medium |
CN110087193A (en) * | 2019-03-05 | 2019-08-02 | 采之翼(北京)科技有限公司 | Information uploading method, device, electronic equipment and the readable storage medium storing program for executing of dustbin |
CN110116415A (en) * | 2019-06-12 | 2019-08-13 | 中北大学 | A kind of Bottle & Can class rubbish identification sorting machine people based on deep learning |
CN110276405A (en) * | 2019-06-26 | 2019-09-24 | 北京百度网讯科技有限公司 | Method and apparatus for output information |
CN110336979A (en) * | 2019-06-24 | 2019-10-15 | 盛皓月 | A kind of mountain forest intelligent garbage monitoring governing system |
WO2020034776A1 (en) * | 2018-08-14 | 2020-02-20 | Oppo广东移动通信有限公司 | Image processing method and apparatus, terminal device, and computer readable storage medium |
CN110852263A (en) * | 2019-11-11 | 2020-02-28 | 北京智能工场科技有限公司 | Mobile phone photographing garbage classification recognition method based on artificial intelligence |
CN111081027A (en) * | 2019-12-17 | 2020-04-28 | 浙江大华技术股份有限公司 | License plate recognition method and device, computer device and readable storage medium |
CN111086796A (en) * | 2019-11-05 | 2020-05-01 | 广东拜登网络技术有限公司 | Perfecting method for garbage type identification and background server |
CN111414926A (en) * | 2019-01-04 | 2020-07-14 | 卓望数码技术(深圳)有限公司 | Intelligent garbage classification method and device, storage medium and robot |
CN111444977A (en) * | 2020-04-03 | 2020-07-24 | 成都禧来科技有限公司 | Method for realizing automatic garbage classification |
CN111754374A (en) * | 2020-06-28 | 2020-10-09 | 北京伟杰东博信息科技有限公司 | Soil pollution monitoring method and system |
WO2021022475A1 (en) * | 2019-08-06 | 2021-02-11 | 中国长城科技集团股份有限公司 | Refuse disposal method and apparatus, and terminal device |
CN112686162A (en) * | 2020-12-31 | 2021-04-20 | 北京每日优鲜电子商务有限公司 | Method, device, equipment and storage medium for detecting clean state of warehouse environment |
CN112707058A (en) * | 2020-12-10 | 2021-04-27 | 广东芯盾微电子科技有限公司 | Detection method, system, device and medium for standard actions of kitchen waste |
CN113139524A (en) * | 2017-12-14 | 2021-07-20 | 创新先进技术有限公司 | Data processing method, device and equipment |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5641072A (en) * | 1994-04-12 | 1997-06-24 | Kurimoto, Ltd. | Method for sorting post-consumed bottles and apparatus thereof |
CN101276417A (en) * | 2008-04-17 | 2008-10-01 | 上海交通大学 | Method for filtering internet cartoon medium rubbish information based on content |
CN103810488A (en) * | 2012-11-09 | 2014-05-21 | 阿里巴巴集团控股有限公司 | Image feature extraction method, image searching method, server, terminal and system |
CN104951784A (en) * | 2015-06-03 | 2015-09-30 | 杨英仓 | Method of detecting absence and coverage of license plate in real time |
CN105787506A (en) * | 2016-01-26 | 2016-07-20 | 耿春茂 | Method for assessing garbage classification based on image identification and two dimensional identification technology |
-
2017
- 2017-02-20 CN CN201710090311.5A patent/CN106874954A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5641072A (en) * | 1994-04-12 | 1997-06-24 | Kurimoto, Ltd. | Method for sorting post-consumed bottles and apparatus thereof |
CN101276417A (en) * | 2008-04-17 | 2008-10-01 | 上海交通大学 | Method for filtering internet cartoon medium rubbish information based on content |
CN103810488A (en) * | 2012-11-09 | 2014-05-21 | 阿里巴巴集团控股有限公司 | Image feature extraction method, image searching method, server, terminal and system |
CN104951784A (en) * | 2015-06-03 | 2015-09-30 | 杨英仓 | Method of detecting absence and coverage of license plate in real time |
CN105787506A (en) * | 2016-01-26 | 2016-07-20 | 耿春茂 | Method for assessing garbage classification based on image identification and two dimensional identification technology |
Non-Patent Citations (1)
Title |
---|
魏书法等: ""基于图像的城市场景垃圾自动检测"", 《集成技术》 * |
Cited By (31)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107767548A (en) * | 2017-08-01 | 2018-03-06 | 山东中公网络科技有限公司 | A kind of intelligent article retracting device, intelligent article recovery system and recovery method |
CN107967460A (en) * | 2017-12-08 | 2018-04-27 | 重庆广睿达科技有限公司 | A kind of rubbish thing based on deep neural network burns recognition methods and system |
CN107967460B (en) * | 2017-12-08 | 2020-05-08 | 重庆广睿达科技有限公司 | Deep neural network-based waste incineration identification method and system |
CN113139524A (en) * | 2017-12-14 | 2021-07-20 | 创新先进技术有限公司 | Data processing method, device and equipment |
CN110008792A (en) * | 2018-01-05 | 2019-07-12 | 比亚迪股份有限公司 | Image detecting method, device, computer equipment and storage medium |
CN108182455A (en) * | 2018-01-18 | 2018-06-19 | 齐鲁工业大学 | A kind of method, apparatus and intelligent garbage bin of the classification of rubbish image intelligent |
CN108960347A (en) * | 2018-03-28 | 2018-12-07 | 中国科学院软件研究所 | A kind of recruitment evaluation system and method for convolutional neural networks handwriting recongnition Ranking Stability |
CN108960347B (en) * | 2018-03-28 | 2021-10-22 | 中国科学院软件研究所 | System and method for evaluating effect of handwriting recognition sequencing stability of convolutional neural network |
CN108288077A (en) * | 2018-04-17 | 2018-07-17 | 天津和或节能科技有限公司 | Grading of old paper device establishes device and method, grading of old paper system and method |
CN108665616A (en) * | 2018-05-15 | 2018-10-16 | 刘晓东 | A kind of recovery method and retracting device of rubbish |
CN110830709A (en) * | 2018-08-14 | 2020-02-21 | Oppo广东移动通信有限公司 | Image processing method and device, terminal device and computer readable storage medium |
WO2020034776A1 (en) * | 2018-08-14 | 2020-02-20 | Oppo广东移动通信有限公司 | Image processing method and apparatus, terminal device, and computer readable storage medium |
CN109308479A (en) * | 2018-09-20 | 2019-02-05 | 云南师范大学 | A kind of automatic classified reclaiming method of Campus Garbage based on deep learning |
CN109308479B (en) * | 2018-09-20 | 2021-07-16 | 云南师范大学 | Deep learning-based campus garbage automatic classification and recovery method |
CN109389161A (en) * | 2018-09-28 | 2019-02-26 | 广州大学 | Rubbish identification evolutionary learning method, apparatus, system and medium based on deep learning |
CN111414926A (en) * | 2019-01-04 | 2020-07-14 | 卓望数码技术(深圳)有限公司 | Intelligent garbage classification method and device, storage medium and robot |
CN110087193A (en) * | 2019-03-05 | 2019-08-02 | 采之翼(北京)科技有限公司 | Information uploading method, device, electronic equipment and the readable storage medium storing program for executing of dustbin |
CN110116415A (en) * | 2019-06-12 | 2019-08-13 | 中北大学 | A kind of Bottle & Can class rubbish identification sorting machine people based on deep learning |
CN110336979A (en) * | 2019-06-24 | 2019-10-15 | 盛皓月 | A kind of mountain forest intelligent garbage monitoring governing system |
CN110276405A (en) * | 2019-06-26 | 2019-09-24 | 北京百度网讯科技有限公司 | Method and apparatus for output information |
WO2021022475A1 (en) * | 2019-08-06 | 2021-02-11 | 中国长城科技集团股份有限公司 | Refuse disposal method and apparatus, and terminal device |
CN111086796A (en) * | 2019-11-05 | 2020-05-01 | 广东拜登网络技术有限公司 | Perfecting method for garbage type identification and background server |
CN110852263A (en) * | 2019-11-11 | 2020-02-28 | 北京智能工场科技有限公司 | Mobile phone photographing garbage classification recognition method based on artificial intelligence |
CN111081027A (en) * | 2019-12-17 | 2020-04-28 | 浙江大华技术股份有限公司 | License plate recognition method and device, computer device and readable storage medium |
CN111444977A (en) * | 2020-04-03 | 2020-07-24 | 成都禧来科技有限公司 | Method for realizing automatic garbage classification |
CN111754374A (en) * | 2020-06-28 | 2020-10-09 | 北京伟杰东博信息科技有限公司 | Soil pollution monitoring method and system |
CN111754374B (en) * | 2020-06-28 | 2021-07-20 | 黑龙江省时代资源环境工程设计有限公司 | Soil pollution monitoring method and system |
CN112707058A (en) * | 2020-12-10 | 2021-04-27 | 广东芯盾微电子科技有限公司 | Detection method, system, device and medium for standard actions of kitchen waste |
CN112707058B (en) * | 2020-12-10 | 2022-04-08 | 广东芯盾微电子科技有限公司 | Detection method, system, device and medium for standard actions of kitchen waste |
CN112686162A (en) * | 2020-12-31 | 2021-04-20 | 北京每日优鲜电子商务有限公司 | Method, device, equipment and storage medium for detecting clean state of warehouse environment |
CN112686162B (en) * | 2020-12-31 | 2023-12-15 | 鄂尔多斯市空港大数据运营有限公司 | Method, device, equipment and storage medium for detecting clean state of warehouse environment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106874954A (en) | The method and relevant apparatus of a kind of acquisition of information | |
Sreelakshmi et al. | Capsule neural networks and visualization for segregation of plastic and non-plastic wastes | |
CN105184312B (en) | A kind of character detecting method and device based on deep learning | |
CN107247956B (en) | Rapid target detection method based on grid judgment | |
CN106919920A (en) | Scene recognition method based on convolution feature and spatial vision bag of words | |
CN107679250A (en) | A kind of multitask layered image search method based on depth own coding convolutional neural networks | |
CN106845510A (en) | Chinese tradition visual culture Symbol Recognition based on depth level Fusion Features | |
CN101901346B (en) | Method for identifying unsuitable content in colour digital image | |
CN104573669A (en) | Image object detection method | |
Gyawali et al. | Comparative analysis of multiple deep CNN models for waste classification | |
CN108171136A (en) | A kind of multitask bayonet vehicle is to scheme to search the system and method for figure | |
CN111368895B (en) | Garbage bag target detection method and detection system in wet garbage | |
CN106815604A (en) | Method for viewing points detecting based on fusion of multi-layer information | |
CN110457982A (en) | A kind of crop disease image-recognizing method based on feature transfer learning | |
CN109886161A (en) | A kind of road traffic index identification method based on possibility cluster and convolutional neural networks | |
CN110458082A (en) | A kind of city management case classification recognition methods | |
CN111186656A (en) | Target garbage classification method and intelligent garbage can | |
CN107403191A (en) | A kind of semi-supervised learning machine sorting technique that transfinites with depth structure | |
CN112733936A (en) | Recyclable garbage classification method based on image recognition | |
CN106971145A (en) | A kind of various visual angles action identification method and device based on extreme learning machine | |
Mookkaiah et al. | Design and development of smart Internet of Things–based solid waste management system using computer vision | |
CN106874825A (en) | The training method of Face datection, detection method and device | |
Pathak et al. | Classification of fruits using convolutional neural network and transfer learning models | |
Shah et al. | A method for waste segregation using convolutional neural networks | |
Arunkumar et al. | An internet of things based waste management system using hybrid machine learning technique |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170620 |
|
RJ01 | Rejection of invention patent application after publication |