CN112633269B - Logo recognition method and system - Google Patents
Logo recognition method and system Download PDFInfo
- Publication number
- CN112633269B CN112633269B CN202011578226.1A CN202011578226A CN112633269B CN 112633269 B CN112633269 B CN 112633269B CN 202011578226 A CN202011578226 A CN 202011578226A CN 112633269 B CN112633269 B CN 112633269B
- Authority
- CN
- China
- Prior art keywords
- meta
- logo
- source domain
- data
- model
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 48
- 238000005070 sampling Methods 0.000 claims abstract description 51
- 230000006870 function Effects 0.000 claims abstract description 43
- 238000012360 testing method Methods 0.000 claims description 34
- 238000012549 training Methods 0.000 claims description 24
- 238000013507 mapping Methods 0.000 claims description 20
- 238000013461 design Methods 0.000 description 13
- 230000008569 process Effects 0.000 description 11
- 238000004891 communication Methods 0.000 description 9
- 238000009826 distribution Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 6
- 230000000007 visual effect Effects 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 238000011161 development Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000012512 characterization method Methods 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 238000004590 computer program Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000013136 deep learning model Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000003190 augmentative effect Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 239000002537 cosmetic Substances 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 239000003550 marker Substances 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 230000000638 stimulation Effects 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 238000013526 transfer learning Methods 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000012850 discrimination method Methods 0.000 description 1
- 238000000802 evaporation-induced self-assembly Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000002427 irreversible effect Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 229910052754 neon Inorganic materials 0.000 description 1
- GKAOGPIIYCISHV-UHFFFAOYSA-N neon atom Chemical compound [Ne] GKAOGPIIYCISHV-UHFFFAOYSA-N 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 230000035807 sensation Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000010200 validation analysis 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
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Multimedia (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Image Analysis (AREA)
Abstract
The application discloses a Logo identification method and system. The Logo identification method comprises the following steps: the selection step comprises the following steps: selecting a target domain and a source domain, and grouping the source domain; iterative steps: sampling the small amount of data of the source domain according to a sampling strategy, calculating a loss function of the source domain model, and updating the source domain model by using the sum of the loss functions in a back propagation way to obtain a trained source domain model; the acquisition step: and obtaining the category of the unknown LOGO in the target domain by using the trained source domain model. According to the Logo recognition method and system, the problem of Logo cross-domain is solved through meta-learning, so that generalization of a model is high, and the model can adapt to a new unknown domain, therefore, the requirement on data size of the new domain is low, and the accuracy of Logo recognition of the new domain is improved.
Description
Technical Field
The application relates to the technical field of Logo recognition, in particular to a Logo recognition method and system.
Background
A great precondition that general deep learning performs well on a single task is that massive data are possessed, better data distribution can drive the deep learning to learn better experience, and the general deep learning has higher data dependence. The meta learning is not optimized for a single task, is how to learn, and is quickly adapted to a new task by capturing the similarity of different tasks, so that the meta learning has higher generalization and can solve the problem of less sample learning. When a domain changes, the performance of the existing deep learning model tends to be significantly degraded because the data distribution of different domains tends to be different, whereas the general deep learning model is only an optimization of the data of the existing domain, and is difficult to adapt to the new domain. When the deep learning model in the prior art faces a new domain, a large amount of data is used for retraining, the requirement on the data amount of the new domain is high, and the generalization of the model is low due to the optimization of the data in a single domain.
Therefore, aiming at the current situation, the invention provides the Logo identification method and the Logo identification system, and solves the problem of Logo cross-domain through meta-learning, so that the generalization of the model is higher, and the model can adapt to a new unknown domain, therefore, the requirement on the data size of the new domain is not high, and the accuracy of Logo identification of the new domain is improved.
Disclosure of Invention
The embodiment of the application provides a Logo identification method and a Logo identification system, which are used for at least solving the problem of subjective factor influence in the related technology.
The invention provides a Logo identification method, which comprises the following steps:
the selection step comprises the following steps: selecting a target domain and a source domain, and grouping the source domain;
iterative steps: sampling the small amount of data of the source domain according to a sampling strategy, calculating a loss function of the source domain model, and updating the source domain model by using the sum of the loss functions in a back propagation way to obtain a trained source domain model;
the acquisition step: and obtaining the category of the unknown LOGO in the target domain by using the trained source domain model.
The Logo identification method includes selecting the Logo of part of industries, selecting a target domain and a source domain from the Logo, and grouping the source domain.
The Logo identification method comprises a meta-training sampling strategy and a meta-testing sampling strategy.
In the Logo recognition method, the iteration step comprises the steps of sampling the small quantity of data according to the meta-training sampling strategy and the meta-testing sampling strategy in each iteration, calculating parameters of each new round of source domain model, calculating the loss function according to the parameters, and updating the source domain model by using the sum total back propagation of the loss function to obtain the trained source domain model.
In the Logo recognition method, the step of obtaining includes that the trained source domain model is used for reasoning data in a library to obtain feature mapping of the data in the library, the trained source domain model is used for reasoning the data of the unknown LOGO on the target domain, and the feature mapping is used for comparing the feature mapping of the data in the library with the feature mapping of the data to obtain the category of the unknown LOGO.
The invention provides a Logo recognition system, which is characterized by being suitable for the Logo recognition method, and comprising the following steps:
the selecting unit: selecting a target domain and a source domain, and grouping the source domain;
iteration unit: sampling the small amount of data of the source domain according to a sampling strategy, calculating a loss function of the source domain model, and updating the source domain model by using the sum of the loss functions in a back propagation way to obtain a trained source domain model;
an acquisition unit: and obtaining the category of the unknown LOGO in the target domain by using the trained source domain model.
In the Logo identification system, the selection unit selects the Logo of part of industries, selects the target domain and the source domain from the Logo, and groups the source domain.
The Logo recognition system comprises a meta-training sampling strategy and a meta-testing sampling strategy.
In the Logo recognition system, the iteration unit samples the small quantity of data according to the meta-training sampling strategy and the meta-testing sampling strategy in each iteration, calculates the parameters of each new round of source domain model, calculates the loss function according to the parameters, and uses the sum of the loss functions to back-propagate and update the source domain model to obtain the trained source domain model.
In the Logo recognition system, the obtaining unit uses the trained source domain model to infer data in a library, and after obtaining feature mapping of the data in the library, uses the trained source domain model to infer the data of the unknown Logo on the target domain, and compares the feature mapping of the data in the library with the feature mapping of the data after the feature mapping of the data, so as to obtain the category of the unknown Logo.
Compared with the related art, the Logo recognition method and system provided by the invention solve the Logo cross-domain problem through meta-learning, so that the generalization of the model is higher and the model can adapt to a new unknown domain, therefore, the requirement on the data size of the new domain is not high, and the accuracy of Logo recognition of the new domain is improved.
The details of one or more embodiments of the application are set forth in the accompanying drawings and the description below to provide a more thorough understanding of the other features, objects, and advantages of the application.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application and do not constitute an undue limitation to the application. In the drawings:
FIG. 1 is a flowchart of a Logo recognition method according to an embodiment of the present application;
FIG. 2 is a source domain packet display diagram according to an embodiment of the present application;
FIG. 3 is a schematic diagram of a Logo recognition system according to the present invention;
fig. 4 is a frame diagram of an electronic device according to an embodiment of the present application.
Wherein, the reference numerals are as follows:
the selecting unit: 51;
iteration unit: 52;
an acquisition unit: 53;
81: a processor;
82: a memory;
83: a communication interface;
80: a bus.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described and illustrated below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application. All other embodiments, which can be made by one of ordinary skill in the art without undue burden on the person of ordinary skill in the art based on the embodiments provided herein, are intended to be within the scope of the present application.
It is apparent that the drawings in the following description are only some examples or embodiments of the present application, and it is possible for those of ordinary skill in the art to apply the present application to other similar situations according to these drawings without inventive effort. Moreover, it should be appreciated that while such a development effort might be complex and lengthy, it would nevertheless be a routine undertaking of design, fabrication, or manufacture for those of ordinary skill having the benefit of this disclosure, and thus should not be construed as having the benefit of this disclosure.
Reference in the specification to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is to be expressly and implicitly understood by those of ordinary skill in the art that the embodiments described herein can be combined with other embodiments without conflict.
Unless defined otherwise, technical or scientific terms used herein should be given the ordinary meaning as understood by one of ordinary skill in the art to which this application belongs. Reference to "a," "an," "the," and similar terms herein do not denote a limitation of quantity, but rather denote the singular or plural. The terms "comprising," "including," "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, article, or apparatus that comprises a list of steps or modules (elements) is not limited to only those steps or elements but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus. The terms "connected," "coupled," and the like in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The term "plurality" as used herein refers to two or more. "and/or" describes an association relationship of an association object, meaning that there may be three relationships, e.g., "a and/or B" may mean: a exists alone, A and B exist together, and B exists alone. The character "/" generally indicates that the context-dependent object is an "or" relationship. The terms "first," "second," "third," and the like, as used herein, are merely distinguishing between similar objects and not representing a particular ordering of objects.
The invention is based on cross-domain Logo recognition of meta-learning and is briefly described below.
Meta-learning (meta-learning) is a model of machine learning, also called how to learn (learning to learn). The primary goal of meta-learning is to design additional learners so that the machine learning system can automatically learn the objective function from the additional learners. The design of the extra learner is the core of meta learning, which involves how to learn "meta knowledge". Unlike traditional machine learning modes, i.e., given a task on which to learn, meta-learning emphasizes abstracting, extracting meta-knowledge from multiple homogeneous tasks, and then applying to new tasks. It can be seen that given a task (dataset), we can always learn an optimal function using the procedure described above. This process is very versatile. It is envisioned that our machine learning process will catch the fly if the number of tasks is very large or if the learning process is very slow. Thus, one idea of being natural is: how do tasks that have been previously learned to be maximally utilized to assist in learning new tasks? Transfer learning is one of the effective thinking. Briefly, the migration learning emphasizes that we have a learned source task, then apply it directly to a target task, and then achieve the learning goal by fine tuning on the target task. This has proven to be an effective way of learning. Meta-learning (also known as learning gto learning) is a very effective learning model. Similar to the goal of transfer learning, meta-learning also emphasizes learning experience from related tasks to aid in learning new tasks. The difference is that meta learning is a more general model, the core of which is the characterization and acquisition of "meta-knowledge". It is understood that this meta-knowledge is a general knowledge that is available over a large class of tasks, and is available through some learning means. It has very powerful characterization capabilities on such tasks and can therefore be generalized over more tasks. To obtain meta-knowledge, meta-learning typically assumes that we can obtain some tasks, which are sampled from the task distribution. We assume that from this task distribution, a single source task can be sampled, denoted as two of which represent a training set and a validation set on one task, respectively. Typically, in meta learning, they are in turn called support sets (support sets) and query sets (query sets). We call the process of learning meta-knowledge the meta-train process, which can be expressed as: wherein parameters representing the meta-knowledge learning process. To verify the effect of meta-knowledge, we define a meta-test procedure: the tasks are sampled from the task distribution to form meta-test data, denoted as. Thus, during meta-test, we can apply learned meta-knowledge to meta-test data to train our real task model: notably, we train their parameters adaptively for each task in the above equation, which completes the generalization process. The basic problems of meta-learning can be divided into three main categories: meta-knowledge representation (meta-representation). The meta-knowledge should be characterized, which answers the most important question of meta-learning, i.e. what to learn. Meta-learner (meta-optimizer). I.e. how we should choose the learning algorithm to optimize after the meta-knowledge characterization, i.e. answer the question how to learn in meta-learning. Meta-target (meta-object). With the knowledge characterization and learning method we should learn towards what targets? This answers the question why the meta-learning is to be so learned.
logo is a foreign language abbreviation of logo or trademark, is an abbreviation of LOGOtype, is a small visual design for identifying identity, plays a role in identifying and popularizing logo-owned companies, and enables consumers to memorize company main bodies and brand cultures through the logo of the image. Logo logos in networks are mainly graphic logos used by various websites to link with other websites, representing one website or one plate of a website. LOGO is a variation of Greek LOGO, a product of modern economies, which is different from ancient imprints. Modern signs carry intangible assets of an enterprise and are a medium for the comprehensive information transfer of the enterprise. The mark is the most important part of the CIS strategy of the enterprise, is the most widely applied element with the highest occurrence frequency in the process of transmitting the enterprise image, and is the most critical element at the same time. The powerful overall strength, perfect management mechanism, high-quality products and services of enterprises are all covered in the mark, and the enterprise is deeply left in the audience through continuous stimulation and repeated depiction. The concept of modern logo is more perfect and mature, and the popularization and application of the logo establish a perfect system. With the advent of the digital age and the rapid development of network culture, traditional information transmission modes and reading modes are challenged unprecedented. The conceptual criteria of efficiency, time are redefined as well, in which case the style of logo also appears to evolve toward individualization, diversification. For the logo creation and designer, a few tens of times more information is expressed by a compact logo than before. Classical logo and designs with foresight and exploratory trends coexist, and the design latitude is expanded. The social and economic measurement standard is not just the quantity of commodity, the performance is good or bad, the existence or nonexistence of category, and the accuracy and speed of concept transmission become the key of new measurement standard and win. It can be said that the era provides an unprecedented practical space for logo creation. Based on this, the comprehensive consideration of the uniqueness and the recognizability, the rationality and the sensibility, the individuality and the commonality of the mark becomes an effective path for the designer to pursue success. Summarizing the current social and market situation, logo can be roughly summarized as the following development trends. Various marks occupy the visual market of the user in a wide market space, and attract customers. Therefore, how to jump out from a plurality of marks is easy to distinguish and remember, and individuality becomes a new requirement. Personalization includes personalization of consumer market demands and personalization from designers. Different consumers have different aesthetic orientations, different merchandise sensations, different designer creatives, and different performances. Thus, personalization is an irreversible trend on multiple platforms, both for the consumer market and for the designer. Since the end of the 19 th century, designs tended to be mechanized due to the effects of the industrial revolution and the style of the Baohaus design, with a feeling of ice cold in the large industrial age. With the development of society and the diversification of aesthetic and human attention, humanization becomes an important factor in design. As noted by the well-known industry designers, design historians, design educators and Proses in the United states, "people always have three dimensions for design: aesthetics, technology, economy, but more importantly, the fourth dimension: humanity-! As well as "logo" should be shaped and shaped according to psychological needs and visual preference. Color and other aspects tend to be humanized, and the color and other aspects are targeted. The characteristics of the information age make the present logo different from the past, and besides showing brand or enterprise attributes, the logo also requires richer visual effects, more vivid shapes, images and color elements which are more suitable for consumption psychology, and the like. Meanwhile, the self-unique design language is translated and created by integrating comprehensive information of multiple aspects of the enterprise, so that the mark can not only vividly and closely express the ideas and the spirit of the enterprise, but also cooperate with the market to perform visual stimulation and attraction on consumers to assist propaganda and sales. The mark becomes a visual link and bridge between the information sender and the information receiver, so that whether the analysis of the information content is accurate or not becomes a way of logo winning. Color and other aspects tend to be humanized, and the color and other aspects are targeted. The artistic expression mode of logo is diversified day by day due to the diversification of conscious forms, namely, a two-dimensional plane form and a common semi-three-dimensional relief concave-convex form; the device is provided with a three-dimensional sign and a dynamic neon sign; a written mark is arranged, and a written mark is also arranged; there are strict markers, as well as conceptual markers. With the advancement of network technology and the development of electronic commerce, network logos are becoming increasingly popular new logo forms. The mark recognition is divided into two main types, namely artificial mark recognition and unmanned mark recognition. The artificial markers function to provide drawing information to the virtual object. The advantage of using a manual sign is that an operator can interact with the virtual object in real time through it. However, in some cases the manual sign may not be used, such as digital repair of a garden. The virtual building is displayed on ruins by using the augmented reality technology, so that the effect of reproducing the garden is achieved. In this case, the marker cannot be placed on the ruins, which can only be achieved by means of an unmanned marker. Artificial logo recognition includes logo region recognition and logo pattern recognition, which are important steps for realizing augmented reality. The process of identifying the mark includes binarizing the image containing the artificial mark, and adopting the algorithm of connected domain extraction to realize the identification of the mark region. There are various methods for identifying different logo patterns, such as: a connected domain quantity discrimination method and a template matching method. The artificial mark is well identified, and the method is an important basis for realizing real-time fusion of virtual and real scenes.
According to the Logo recognition method and system, the problem of Logo cross-domain is solved through meta-learning, so that generalization of a model is high, and the model can adapt to a new unknown domain, therefore, the requirement on data size of the new domain is low, and the accuracy of Logo recognition of the new domain is improved.
Next, logo is identified as an example to describe an embodiment of the present application.
Example 1
The embodiment provides a Logo recognition method. Referring to fig. 1-2, fig. 1 is a flowchart of a Logo recognition method according to an embodiment of the application; fig. 2 is a source domain packet display diagram according to an embodiment of the present application, as shown in the drawing, a Logo identification method includes the following steps:
selecting step S1: selecting a target domain and a source domain, and grouping the source domain;
iterative step S2: sampling the small amount of data of the source domain according to a sampling strategy, calculating a loss function of the source domain model, and updating the source domain model by using the sum of the loss functions in a back propagation way to obtain a trained source domain model;
an acquisition step S3: and obtaining the category of the unknown LOGO in the target domain by using the trained source domain model.
In an embodiment, the selecting step S1 includes selecting a Logo of a part of industries, selecting a target domain and a source domain from the Logo, and grouping the source domain.
In specific implementation, part of industry LOGO data is selected, and the industry comprises: cosmetic, automotive, food, electronic, sports, apparel, luxury. The meta-learning scheme is illustrated by taking cosmetic, automobile, food, electronics, sports, clothing as a source domain and luxury as a target domain as an example, and other industries can be used as the target domain, and the rest industries are used as the source domain for model training and testing. Industries of the source domain may be divided into 6 groups, each group being used by K-1 industries for meta-training and 1 industry for meta-testing, as shown in FIG. 2. Group 1 is denoted by D1, group D2 by D2, and so on.
In an embodiment, the sampling strategy includes a meta-training sampling strategy and a meta-testing sampling strategy.
In a specific implementation, the meta-training sampling strategy is: from K-1 industries, M are selected from each industry, N samples are selected from each category, and then the small data size is (K-1) M N. The meta-test sampling strategy is: from 1 industry, M industries are selected, N samples are selected for each category, and then the small data size is M.
In an embodiment, the step S2 of iterating includes, in each iteration, sampling the small number of data according to the meta-training sampling strategy and the meta-testing sampling strategy, calculating parameters of each new source domain model, calculating the loss function according to the parameters, and updating the source domain model by using sum back propagation of the loss function to obtain the trained source domain model.
In the specific implementation, each iteration, the meta-training stage only calculates new model parameters according to the loss function, and the model is not updated by back propagation. Meta-test stage computing contrast and domain alignment loss function L on new model parameters (obtained in meta-training stage) meta-test . Calculating one per packetL meta-test Then, sum all packets Σl meta-test And back-propagates the update model using this loss function. The loss function comprises two parts: comparison (L) contrastiv e ) Domain alignment (L) align ) The intra-class spacing is reduced by using a loss function of metric learning, and the inter-class spacing is increased; the influence of the domain on LOGO characteristic distribution is reduced by using the domain alignment loss function, so that the LOGO characteristic distribution is distributed according to the LOGO characteristics.
In an embodiment, the step S3 of obtaining includes using the trained source domain model to infer data in a library, obtaining feature mapping of the data in the library, using the trained source domain model to infer the data of the unknown LOGO on the target domain, and comparing the feature mapping of the data in the library with the feature mapping of the data after the feature mapping of the data, to obtain the category of the unknown LOGO. The pseudo code of iterative step S2 is as follows:
in the specific implementation, the authentication service is called to perform unified authority management, the authority abnormality directly returns related information to the user side, the authority is normal, SQL portrait information is constructed through the SQL portrait module, related calculation engines are called through the routing module, and the results are combined to the client side for returning after calculation of different engines is completed.
Therefore, the Logo recognition method and system provided by the invention solve the problem of Logo cross-domain through meta-learning, so that the generalization of the model is higher and the model can adapt to a new unknown domain, the requirement on the data size of the new domain is low, and the accuracy of Logo recognition of the new domain is improved.
Example two
Referring to fig. 3, fig. 3 is a schematic structural diagram of the Logo recognition system of the present invention. As shown in fig. 3, the Logo recognition system of the present invention is applicable to the Logo recognition method described above, and the Logo recognition system includes:
the selecting unit 51: selecting a target domain and a source domain, and grouping the source domain;
iteration unit 52: sampling the small amount of data of the source domain according to a sampling strategy, calculating a loss function of the source domain model, and updating the source domain model by using the sum of the loss functions in a back propagation way to obtain a trained source domain model;
the acquisition unit 53: and obtaining the category of the unknown LOGO in the target domain by using the trained source domain model.
In this embodiment, the selecting unit 51 selects a Logo of a part of industries, selects a target domain and a source domain from the logos, and groups the source domains.
In this embodiment, the sampling strategies include a meta-training sampling strategy and a meta-testing sampling strategy.
In this embodiment, in each iteration, the iteration unit 52 samples the small amount of data according to the meta-training sampling strategy and the meta-testing sampling strategy, calculates parameters of each new source domain model, calculates the loss function according to the parameters, and uses the sum of the loss functions to back-propagate and update the source domain model to obtain the trained source domain model.
In this embodiment, the obtaining unit 53 uses the trained source domain model to infer data in a library, and after obtaining a feature map of the data in the library, uses the trained source domain model to infer the data of the unknown LOGO on the target domain, and compares the feature map of the data in the library with the feature map of the data after performing the feature map to obtain the category of the unknown LOGO.
Example III
In connection with fig. 4, this embodiment discloses a specific implementation of an electronic device. The electronic device may include a processor 81 and a memory 82 storing computer program instructions.
In particular, the processor 81 may include a Central Processing Unit (CPU), or an application specific integrated circuit (Application Specific Integrated Circuit, abbreviated as ASIC), or may be configured to implement one or more integrated circuits of embodiments of the present application.
Memory 82 may include, among other things, mass storage for data or instructions. By way of example, and not limitation, memory 82 may comprise a Hard Disk Drive (HDD), floppy Disk Drive, solid state Drive (Solid State Drive, SSD), flash memory, optical Disk, magneto-optical Disk, tape, or universal serial bus (Universal Serial Bus, USB) Drive, or a combination of two or more of the foregoing. The memory 82 may include removable or non-removable (or fixed) media, where appropriate. The memory 82 may be internal or external to the data processing apparatus, where appropriate. In a particular embodiment, the memory 82 is a Non-Volatile (Non-Volatile) memory. In a particular embodiment, the Memory 82 includes Read-Only Memory (ROM) and random access Memory (Random Access Memory, RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (Programmable Read-Only Memory, abbreviated PROM), an erasable PROM (Erasable Programmable Read-Only Memory, abbreviated FPROM), an electrically erasable PROM (Electrically Erasable Programmable Read-Only Memory, abbreviated EFPROM), an electrically rewritable ROM (Electrically Alterable Read-Only Memory, abbreviated EAROM), or a FLASH Memory (FLASH), or a combination of two or more of these. The RAM may be Static Random-Access Memory (SRAM) or dynamic Random-Access Memory (Dynamic Random Access Memory DRAM), where the DRAM may be a fast page mode dynamic Random-Access Memory (Fast Page Mode Dynamic Random Access Memory FPMDRAM), extended data output dynamic Random-Access Memory (Extended Date Out Dynamic Random Access Memory EDODRAM), synchronous dynamic Random-Access Memory (Synchronous Dynamic Random-Access Memory SDRAM), or the like, as appropriate.
Memory 82 may be used to store or cache various data files that need to be processed and/or communicated, as well as possible computer program instructions for execution by processor 81.
The processor 81 reads and executes the computer program instructions stored in the memory 82 to implement any Logo recognition method in the above embodiment.
In some of these embodiments, the electronic device may also include a communication interface 83 and a bus 80. As shown in fig. 4, the processor 81, the memory 82, and the communication interface 83 are connected to each other through the bus 80 and perform communication with each other.
The communication interface 83 is used to implement communications between various modules, devices, units, and/or units in embodiments of the present application. Communication port 83 may also enable communication with other components such as: and the external equipment, the image/data acquisition equipment, the database, the external storage, the image/data processing workstation and the like are used for data communication.
Bus 80 includes hardware, software, or both that couple components of the electronic device to one another. Bus 80 includes, but is not limited to, at least one of: data Bus (Data Bus), address Bus (Address Bus), control Bus (Control Bus), expansion Bus (Expansion Bus), local Bus (Local Bus). By way of example, and not limitation, bus 80 may include a graphics acceleration interface (Accelerated Graphics Port), abbreviated AGP, or other graphics Bus, an enhanced industry standard architecture (Extended Industry Standard Architecture, abbreviated EISA) Bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an industry standard architecture (Industry Standard Architecture, ISA) Bus, a wireless bandwidth (InfiniBand) interconnect, a Low Pin Count (LPC) Bus, a memory Bus, a micro channel architecture (Micro Channel Architecture, abbreviated MCa) Bus, a peripheral component interconnect (Peripheral Component Interconnect, abbreviated PCI) Bus, a PCI-Express (PCI-X) Bus, a serial advanced technology attachment (Serial Advanced Technology Attachment, abbreviated SATA) Bus, a video electronics standards association local (Video Electronics Standards Association Local Bus, abbreviated VLB) Bus, or other suitable Bus, or a combination of two or more of the foregoing. Bus 80 may include one or more buses, where appropriate. Although embodiments of the present application describe and illustrate a particular bus, the present application contemplates any suitable bus or interconnect.
The electronic device may be connected to the Logo recognition system to implement the method described in connection with fig. 1-2.
The technical features of the above-described embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above-described embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples merely represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the invention. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application is to be determined by the claims appended hereto.
Claims (6)
1. The Logo recognition method is characterized by comprising the following steps of:
the selection step comprises the following steps: selecting a target domain and a source domain, grouping the source domain, namely selecting LOGO data of part of industries, wherein each group is used for meta-training by K-1 industries, and performing meta-test by the other 1 industry;
iterative steps: sampling the small amount of data of the source domain according to a sampling strategy, calculating a loss function of the source domain model, and updating the source domain model by using the sum of the loss functions in a back propagation way to obtain a trained source domain model;
the acquisition step: obtaining the category of the unknown LOGO in the target domain by using the trained source domain model;
the sampling strategy comprises a meta training sampling strategy and a meta testing sampling strategy, the iteration step comprises the steps of sampling the small quantity of data according to the meta training sampling strategy and the meta testing sampling strategy in each iteration, calculating parameters of each new round of source domain model, calculating the loss function according to the parameters, and updating the source domain model by using the sum back propagation of the loss function to obtain the trained source domain model;
wherein the meta-training sampling strategy is: selecting M industries from K-1 industries, and selecting N samples from each category, wherein the size of the small data is (K-1) M N; the meta-test sampling strategy is: selecting M classes from 1 industry, wherein each class selects N samples, and the size of the small data is M x N; for each iteration, the meta-training stage only calculates new model parameters according to the loss function, and does not update the model by back propagation; the meta-test stage calculates the contrast and domain alignment loss function L on new model parameters meta-test The method comprises the steps of carrying out a first treatment on the surface of the Calculating one L per packet meta-test Then, sum all the packets, Σl meta-test And back-propagating the update model using this loss function; the loss function comprises two parts: contrast and domain alignment.
2. The Logo recognition method as claimed in claim 1, wherein the selecting step includes selecting a Logo of a part of industries, selecting a target domain and a source domain from the Logo, and grouping the source domains.
3. The Logo recognition method as claimed in claim 1, wherein the obtaining step includes using the trained source domain model to infer data in a library, obtaining feature mapping of the data in the library, using the trained source domain model to infer the data of the unknown Logo on the target domain, and comparing the feature mapping of the data in the library with the feature mapping of the data after the feature mapping of the data, to obtain the category of the unknown Logo.
4. A Logo recognition system, which is suitable for the Logo recognition method as claimed in claims 1-3, and comprises:
the selecting unit: selecting a target domain and a source domain, grouping the source domain, namely selecting LOGO data of part of industries, wherein each group is used for meta-training by K-1 industries, and performing meta-test by the other 1 industry;
iteration unit: sampling the small amount of data of the source domain according to a sampling strategy, calculating a loss function of the source domain model, and updating the source domain model by using the sum of the loss functions in a back propagation way to obtain a trained source domain model;
an acquisition unit: obtaining the category of the unknown LOGO in the target domain by using the trained source domain model;
the sampling strategy comprises a meta training sampling strategy and a meta testing sampling strategy, the iteration unit samples the small quantity of data according to the meta training sampling strategy and the meta testing sampling strategy in each iteration, calculates parameters of each new round of source domain model, calculates the loss function according to the parameters, and uses the sum of the loss functions to back propagate and update the source domain model to obtain the trained source domain model;
wherein the meta-training sampling strategy is: selecting M industries from K-1 industries, and selecting N samples from each category, wherein the size of the small data is (K-1) M N; the meta-test sampling strategy is: selecting M classes from 1 industry, wherein each class selects N samples, and the size of the small data is M x N; for each iteration, the meta-training stage only calculates new model parameters according to the loss function, and does not update the model by back propagation; the meta-test stage calculates the contrast and domain alignment loss function L on new model parameters meta-test The method comprises the steps of carrying out a first treatment on the surface of the Calculating one L per packet meta-test Then, sum all the packets, Σl meta-test And back-propagating the update model using this loss function; the loss function comprises two parts: contrast and domain alignment.
5. The Logo recognition system as claimed in claim 4, wherein the selection unit selects a Logo of a part of industries, selects a target domain and a source domain from the Logo, and groups the source domains.
6. The Logo recognition system as claimed in claim 4, wherein the obtaining unit uses the trained source domain model to infer data in a library, obtains feature mapping of the data in the library, uses the trained source domain model to infer the data of the unknown Logo on the target domain, and compares the feature mapping of the data in the library with the feature mapping of the data after the feature mapping of the data, so as to obtain the category of the unknown Logo.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011578226.1A CN112633269B (en) | 2020-12-28 | 2020-12-28 | Logo recognition method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011578226.1A CN112633269B (en) | 2020-12-28 | 2020-12-28 | Logo recognition method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112633269A CN112633269A (en) | 2021-04-09 |
| CN112633269B true CN112633269B (en) | 2024-03-15 |
Family
ID=75326036
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011578226.1A Active CN112633269B (en) | 2020-12-28 | 2020-12-28 | Logo recognition method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112633269B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113569934B (en) * | 2021-07-20 | 2024-01-23 | 上海明略人工智能(集团)有限公司 | LOGO classification model construction method, LOGO classification model construction system, electronic equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106934392A (en) * | 2017-02-28 | 2017-07-07 | 西交利物浦大学 | Vehicle-logo recognition and attribute forecast method based on multi-task learning convolutional neural networks |
| CN108898218A (en) * | 2018-05-24 | 2018-11-27 | 阿里巴巴集团控股有限公司 | A kind of training method of neural network model, device and computer equipment |
| CN110472667A (en) * | 2019-07-19 | 2019-11-19 | 广东工业大学 | Small object classification method based on deconvolution neural network |
| CN110991506A (en) * | 2019-11-22 | 2020-04-10 | 高新兴科技集团股份有限公司 | Vehicle brand identification method, device, equipment and storage medium |
| CN111721536A (en) * | 2020-07-20 | 2020-09-29 | 哈尔滨理工大学 | Rolling bearing fault diagnosis method for improving model migration strategy |
-
2020
- 2020-12-28 CN CN202011578226.1A patent/CN112633269B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106934392A (en) * | 2017-02-28 | 2017-07-07 | 西交利物浦大学 | Vehicle-logo recognition and attribute forecast method based on multi-task learning convolutional neural networks |
| CN108898218A (en) * | 2018-05-24 | 2018-11-27 | 阿里巴巴集团控股有限公司 | A kind of training method of neural network model, device and computer equipment |
| CN110472667A (en) * | 2019-07-19 | 2019-11-19 | 广东工业大学 | Small object classification method based on deconvolution neural network |
| CN110991506A (en) * | 2019-11-22 | 2020-04-10 | 高新兴科技集团股份有限公司 | Vehicle brand identification method, device, equipment and storage medium |
| CN111721536A (en) * | 2020-07-20 | 2020-09-29 | 哈尔滨理工大学 | Rolling bearing fault diagnosis method for improving model migration strategy |
Non-Patent Citations (2)
| Title |
|---|
| Learning Meta Face Recognition in Unseen Domains;Jianzhu Guo等;《https://arxiv.org/abs/2003.07733》;论文第1-10页 * |
| 元学习（Meta-learning）——李宏毅老师教学视频笔记;思座の机器学习笔记;<https://zhuanlan.zhihu.com/p/108503451>;论文第1-7页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112633269A (en) | 2021-04-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109618222B (en) | A kind of splicing video generation method, device, terminal device and storage medium | |
| US20230368505A1 (en) | Model training method, image retrieval method, and apparatus | |
| Matz et al. | Predicting the personal appeal of marketing images using computational methods | |
| CN111046275B (en) | User label determining method and device based on artificial intelligence and storage medium | |
| CN113254549B (en) | Character relation mining model training method, character relation mining method and device | |
| CN115131698B (en) | Video attribute determining method, device, equipment and storage medium | |
| CN112633269B (en) | Logo recognition method and system | |
| CN114972847A (en) | Image processing method and device | |
| CN116089645A (en) | Hierarchical style-based conditional text-e-commerce picture retrieval method and system | |
| CN112767038B (en) | Poster CTR prediction method and device based on aesthetic characteristics | |
| CN112100493B (en) | Document ordering method, device, equipment and storage medium | |
| CN117994570A (en) | Recognition method for improving complex and diverse data distribution based on model-independent adapter | |
| CN111191059B (en) | Image processing method, device, computer storage medium and electronic equipment | |
| CN116756281A (en) | Knowledge question-answering method, device, equipment and medium | |
| CN109299378B (en) | Search result display method and device, terminal and storage medium | |
| Wang | Wireless multifunctional display platform for visual communication design based on iot big data | |
| CN111259975A (en) | Method and device for generating classifier and method and device for classifying text | |
| Wang et al. | [Retracted] Visual Communication Design Model for New Media and Public Health Environment and New Communication Mode | |
| CN115115869A (en) | Business image labeling method and device, electronic equipment and computer program product | |
| CN112328799B (en) | Question classification method and device | |
| CN108460131A (en) | A kind of tag along sort processing method and processing device | |
| Qin et al. | Statistical System of Cultural Heritage Tourism Information Based on Image Feature Extraction Technology | |
| CN108830298A (en) | A kind of method and device of determining user characteristics label | |
| CN115035343A (en) | Neural network training method, article detection method, apparatus, device and medium | |
| CN113536111A (en) | Insurance knowledge content recommendation method and device and terminal equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |