EP3948664A1 - Procede et systeme d'identification d'objets a partir d'images labellisees desdits objets - Google Patents
Procede et systeme d'identification d'objets a partir d'images labellisees desdits objetsInfo
- Publication number
- EP3948664A1 EP3948664A1 EP20712507.1A EP20712507A EP3948664A1 EP 3948664 A1 EP3948664 A1 EP 3948664A1 EP 20712507 A EP20712507 A EP 20712507A EP 3948664 A1 EP3948664 A1 EP 3948664A1
- Authority
- EP
- European Patent Office
- Prior art keywords
- labeled
- images
- module
- invariants
- precision
- 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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/778—Active pattern-learning, e.g. online learning of image or video features
- G06V10/7784—Active pattern-learning, e.g. online learning of image or video features based on feedback from supervisors
- G06V10/7788—Active pattern-learning, e.g. online learning of image or video features based on feedback from supervisors the supervisor being a human, e.g. interactive learning with a human teacher
Definitions
- the field of the invention is that of the identification of objects from labeled images (labeled) of said objects.
- the term “labeled image” of a wood defect is understood to mean an image in which a defect has been linked to a known defect belonging to the group formed for example by crack, termite hole, shock hole.
- machine learning or “Machine Learning” in Anglo-Saxon, which works by automatic learning of a base of labeled invariants (attributes or characteristics).
- An algorithm is trained with a large amount of input data resulting in the creation of a model that allows output to be provided.
- the model here is a representation of the question asked and how to answer it.
- the second machine learning approach requires a representative, not necessarily factorial, labeled database of labeled invariants in order to be able to recognize or identify a given object.
- the success rate of this approach rarely or slightly exceeds that of the business expert.
- the third approach by Deep Learning needs large labeled 2D / 3D databases, obtained in factorial. She succeeds in identifying a given object with a success rate or accuracy greater than 95%, hence the interest of this approach.
- the aim of the invention is to overcome the drawbacks of the identification approaches of the prior art.
- said first identification module having a first identification precision corresponding to that of the business expert
- a second identification module suitable for receiving as input the images of the object to be identified and the invariants labeled by the business expert and capable of delivering as output a second plurality of images of said object labeled by machine learning and a second plurality of invariants labeled by machine learning; said second identification module having a second precision corresponding to that of machine learning;
- a third identification module suitable for receiving as input the images of the object to be identified and capable of delivering as output a third plurality of images labeled by deep machine learning and a third plurality of invariants labeled by machine learning, said third identification module having a third precision corresponding to that of
- the system further comprises:
- - a module for the approval and pooling of labeled images receiving as input the first, second, and third pluralities of images labeled respectively by the business expert, by machine learning and by deep machine learning and delivering a plurality of 'mutualized labeled images for adjustment with the best precision;
- an aggregation and pooling module of labeled invariants receiving as input the first, second, and third labeled invariant pluralities
- the first, second, and third identification modules use as input for their respective identification, the plurality of labeled pooled adjustment images and / or the plurality of pooled labeled adjustment invariants resulting from the aggregation and pooling modules.
- the identification system benefits not only from the result at the output of the identification module exhibiting the best precision in the learning phase for the identification of a chosen object but also from reusing (pooling ) the best input identification result for the test phase (interrogation) on subsequent object identifications to be processed.
- the identification system in accordance with the invention makes it possible to remedy the drawbacks of each individual approach, to reduce the discrepancies between the updating of the databases, or even between the knowledge extracted and increases the precision of each approach. respective thanks to the pooling of databases (2D / 3D images, 2D / 3D invariants coupled with the best precision) enriched by the best response.
- the best mutualized precision is equal to the maximum between the precision of the Business Expert, the precision of Machine Learning and the precision of Deep Learning.
- the mutualized adjustment label is equal to the Professional Expert label if the Professional Expert precision is greater than the Machine Learning precision and the Deep Learning precision
- the mutualized label of the object is equal the Machine Learning Label if the Machine Learning Precision is greater than the Deep Learning Precision and the Business Expert Precision
- the shared Label is equal for the Deep Learning Label if the Deep Learning Precision is greater than the Expertise Precision and the Machine Learning Precision.
- the mutualized labeled images of adjustment are obtained by consolidating the images of the object with the mutualized label of the object.
- the mutualized labeled adjustment invariants are obtained by consolidating the aggregated Invariants with the mutualized Label of the object, the aggregated Invariants are obtained by the aggregation of the
- Invariants labeled Business Expert Invariants labeled Machine Learning and Invariants labeled Deep Learning.
- the present invention also relates to a method of identifying an object from labeled images of said object implemented by said identification system according to the invention.
- the present invention also relates to a computer program downloadable from a communication network and / or recorded on a computer readable medium and / or executable by a processor, which comprises instructions for the execution of the conforming method to the invention when said program is executed on a computer.
- FIG 1 schematically represents the aggregation and pooling of the results of three approaches: Business Expert, Machine Learning and Deep Learning, [0033]
- FIG 2 schematically represents the process of identifying objects by the Business expert using the shared database of labeled 2D / 3D invariants,
- FIG. 3 schematically represents the complete process of learning and identifying objects by Machine Learning using the shared database of labeled 2D / 3D invariants
- FIG 4 schematically represents the complete process of learning and identifying objects by Deep Learning using the shared database of labeled 2D / 3D images
- FIG. 1 a, b, c, d are illustrations of 2D images of the defects to be identified and the associated labels
- FIG 6 is a label obtained by Business Expert approach
- FIG 7 is a label obtained by Deep Learning approach.
- the system comprises a first IDEM identification module driven by a business expert EM who receives as input IMO images of an object to be identified (2D and / or 3D), IVO invariants extracted by a EXTIV extraction algorithm.
- the first IDEM identification module delivers ILEM labeled images of the object by the business expert (human) according to predefined rules linked to the object to be identified and to the expert's profession, invariants labeled by the IVLEM business expert, and a PEM precision from the EM business expert.
- the hole belongs to the list of specific labels in the wood industry.
- the hole (figure 6) in the beam was identified by the business expert by the label "Hole t_1".
- the identification system comprises a second IDLM identification module driven by machine learning also called Machine Learning which receives as input IMO images of an object to be identified (2D and / or 3D), invariants (geometric or others) and IVO invariants extracted by
- the second IDLM identification module delivers ILML-labeled images of the object by machine learning, IVLML-labeled invariants, and PML precision of machine learning.
- the identification system also further comprises a third IDDL identification module driven by deep machine learning or "Deep Learning” which receives as input IMO images of an object to be identified (2D and / or 3D).
- the third IDDL identification module delivers ILDL labeled images of the object by Deep Learning, IVLDL labeled invariants, and PDL precision of Deep Learning.
- the third IDDL identification module has PDL accuracy corresponding to that of deep machine learning.
- the MAMR module has as inputs:
- the MAMR module generates:
- the PMIM and PMIV mutualized Precisions are chosen so as to be equal at most between the Precision of the PEM Business Expert, the Precision of the PML Machine Learning and the Precision of the Deep Learning PDL.
- the aggregation and pooling module outputs a result which corresponds to the label selected, also called "Label LAB" of the shared object.
- the shared labeled 2D / 3D images are obtained by consolidating the 2D / 3D images of the object with the shared LAB label of the object.
- the shared labeled 2D / 3D Invariants are obtained by consolidating the aggregated Invariants with the shared LAB Label of the object.
- Aggregated 2D / 3D Invariants are obtained by aggregating Invariants labeled by the Business Expert, Invariants labeled by Machine Learning and Invariants labeled by Deep Learning.
- the shared labeled 2D / 3D images and the associated shared Precisions will enrich a shared database of 2D / 3D images labeled BMIM with associated PMIM Precisions.
- pooled labeled 2D / 3D Invariants and pooled Precisions will enrich a pooled database of BMIV labeled 2D / 3D invariants and with associated PMIV Precisions.
- the shared base of 2D / 3D images labeled BMIM with associated details PMIM will be used for the adjustment of the IDLM and IDDL identification modules (dotted in Figure 1).
- the shared database of 2D / 3D invariants labeled BMIV and with associated details PMIV will be used by the IDEM module (dotted in Figure 1).
- BMIM images
- BMIV invariants
- 2D / 3D IMO images of an object to be identified are sent:
- the EXTIV 2D / 3D Invariant Extraction Algorithmic Module generates the 2D / 3D IVO Invariants of the object from the 2D / 3D IMO Images of the object.
- the 2D / 3D IVO Invariants of the object are sent:
- the complete process of identifying objects by the Business Expert will call on the shared database BMIV-labeled 2D / 3 invariants to adjust its human identification if the invariants stored in the database have better precision than those from the EXTIV extraction module.
- the 2D / 3D Invariant Extraction Algorithmic Module EXTIV generates 2D / 3D Invariants of the object from the 2D / 3D Images of the object.
- the search is positive, that is to say when the labeled 2D / 3D Invariants resulting from the mutualized Base of 2D / 3D invariants BMIV are better than those resulting from the extraction module, then the labeled invariants with associated details from the BMIV database are in turn sent to the Business Expert as an aid in the human identification of the object. Otherwise, in the event of a negative search, the expert relies on the invariants from the EXTIV extraction module.
- the Business Expert generates the ILEM Label of the object and the associated PEM Precision on the basis of the invariants thus updated and optimized thanks to the database of labeled and shared invariants.
- the CREM results consolidation algorithmic module generates the 2D / 3D Images of the object labeled by the Business Expert, the 2D / 3D Invariants of the object labeled by the Business Expert and the Accuracy of the Expert Business by consolidating
- the process includes a Machine Learning phase in order to adjust (optimize, update) the Machine Model
- the process includes a questioning phase of the Machine Learning Model IDML, adjusted during the phase
- the 2D / 3D Invariant Extraction Algorithmic Module generates 2D / 3D Invariants of the object from 2D / 3D Images of the object
- the results Consolidation algorithmic module generates 2D / 3D Images of the object labeled by Machine Learning, 2D / 3D Invariants of the object labeled by Machine Learning and Machine Learning Precision by consolidating (associating) the Label of the object to the 2D / 3D Images of the object, to the 2D / 3D Invariants of the object and to the associated Precision.
- a Deep Learning Learning phase makes it possible to adjust the Deep Learning Model on the correspondence 2D / 3D Images x Labels from the shared database of labeled 2D / 3D images with associated details.
- the process includes a questioning phase of the Deep Learning IDDL Model, adjusted during the phase
- the Interrogation Module of the IDDL adjusted Deep Learning Model generates both the Label of the object, the associated Precision and the 2D / 3D Invariants of the object,
- the algorithmic module for Consolidation of CRDL results generates the 2D / 3D Images of the object labeled by Deep Learning, the 2D / 3D Invariants of the object labeled by Deep Learning and the Precision of Deep Learning by consolidating (associating) the Label of the object to the 2D / 3D Images of the object, to the 2D / 3D Invariants of the object and to the associated Precision.
- Figures 5a to 5d there is illustrated an example of four images of a wooden beam comprising anomalies or defects to be identified.
- the problem here is to identify the crippling defect in the wood industry on a large quantity of wooden beams to be treated.
- the hole belongs to the list of specific labels in the wood industry.
- the hole (figure 6) in the beam has been identified by business expert and machine learning and labeled with the label “Hole t_1”.
- f2Joc location of Fissure_2 [x1, y1, x2, y2]
- tm2Joc location holes_ termites_2 [x1, y1]
- tm3_loc location holes_ termites_3 [x1, y1]
- tm5_loc location holes_ termites_5 [x1, y1]
- tm7_loc location holes_ termites_7 [x1, y1]
- b_mony condition of the wood (good, average, bad).
- Labels obtained by the Deep Learning Approach and the regions of interest associated with anomalies or defects on the same image of the wooden beam. These Labels are of the type: Hole, Fissure and Termite Hole.
- the structure of the Machine Learning Approach used for the classification of defects is of the Support Vector Machine (SVM) type.
- SVM Support Vector Machine
- the structure of the Deep Learning Approach used for the detection and classification of defects is of the "Faster R-CNN (Regions with Convolutional Neural Network features) object classification” type.
- the aggregation of invariants uses adaptive Boolean operators: AND, OR and XOR.
- Other methods are applicable: Bayesian probabilistic methods, the Dempster-Shafer method ”based on belief theory, Borda Count rank methods, Ordered Weighted Averaging operator (OWA), Aggregator Weight-Functional Operator (AWFO) , Hidden Markov Chains (CMC), rule-based fuzzy logic
- the invention thus makes it possible to remedy the drawbacks of each individual approach, to reduce the discrepancies between the updating of the databases, or even between the knowledge extracted, and increases the details of each respective approach thanks to the pooling of the databases.
- the Deep Learning approach uses the Labels acquired in learning by the Approach Business Expert and Machine Learning Approach Labels,
- the Machine Learning Approach uses the invariants acquired in learning by the Business Expert Approach and the invariants of the Deep Learning Approach,
- the identification system benefits not only from the result at the output of the identification module having the best precision in the learning phase for the identification of a chosen object but also to reuse
- the fields of application of the invention are broad, covering the detection, classification, recognition and identification of objects of interest.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
Description
Claims
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
FR1903232A FR3094532B1 (fr) | 2019-03-28 | 2019-03-28 | Procédé et système d'identification d'objets à partir d'images labellisées desdits objets |
PCT/EP2020/057107 WO2020193253A1 (fr) | 2019-03-28 | 2020-03-16 | Procede et systeme d'identification d'objets a partir d'images labellisees desdits objets |
Publications (1)
Publication Number | Publication Date |
---|---|
EP3948664A1 true EP3948664A1 (fr) | 2022-02-09 |
Family
ID=68072534
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
EP20712507.1A Pending EP3948664A1 (fr) | 2019-03-28 | 2020-03-16 | Procede et systeme d'identification d'objets a partir d'images labellisees desdits objets |
Country Status (6)
Country | Link |
---|---|
US (1) | US20220180620A1 (fr) |
EP (1) | EP3948664A1 (fr) |
JP (1) | JP2022525812A (fr) |
KR (1) | KR20210140740A (fr) |
FR (1) | FR3094532B1 (fr) |
WO (1) | WO2020193253A1 (fr) |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP4253522B2 (ja) * | 2003-03-28 | 2009-04-15 | 株式会社日立ハイテクノロジーズ | 欠陥分類方法及び装置 |
FR2981772B1 (fr) | 2011-10-21 | 2017-12-22 | Thales Sa | Procede de reconstruction 3d d'un objet d'une scene |
US10095917B2 (en) | 2013-11-04 | 2018-10-09 | Facebook, Inc. | Systems and methods for facial representation |
-
2019
- 2019-03-28 FR FR1903232A patent/FR3094532B1/fr active Active
-
2020
- 2020-03-16 EP EP20712507.1A patent/EP3948664A1/fr active Pending
- 2020-03-16 US US17/598,177 patent/US20220180620A1/en active Pending
- 2020-03-16 JP JP2021557475A patent/JP2022525812A/ja active Pending
- 2020-03-16 KR KR1020217032215A patent/KR20210140740A/ko unknown
- 2020-03-16 WO PCT/EP2020/057107 patent/WO2020193253A1/fr unknown
Also Published As
Publication number | Publication date |
---|---|
KR20210140740A (ko) | 2021-11-23 |
JP2022525812A (ja) | 2022-05-19 |
FR3094532A1 (fr) | 2020-10-02 |
WO2020193253A1 (fr) | 2020-10-01 |
FR3094532B1 (fr) | 2021-09-10 |
US20220180620A1 (en) | 2022-06-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Rezende et al. | Malicious software classification using transfer learning of resnet-50 deep neural network | |
WO2022121289A1 (fr) | Procédés et systèmes d'exploitation d'échantillons de données de classe minoritaire destinés à la formation d'un réseau de neurones | |
US10719780B2 (en) | Efficient machine learning method | |
Prajwala | A comparative study on decision tree and random forest using R tool | |
US10650508B2 (en) | Automatic defect classification without sampling and feature selection | |
US20200143209A1 (en) | Task dependent adaptive metric for classifying pieces of data | |
Salunkhe et al. | A hybrid approach for class imbalance problem in customer churn prediction: A novel extension to under-sampling | |
CN112487406B (zh) | 一种基于机器学习的网络行为分析方法 | |
CN114092742A (zh) | 一种基于多角度的小样本图像分类装置和方法 | |
CN111310820A (zh) | 基于交叉验证深度cnn特征集成的地基气象云图分类方法 | |
US11397868B2 (en) | Fungal identification by pattern recognition | |
WO2020193253A1 (fr) | Procede et systeme d'identification d'objets a partir d'images labellisees desdits objets | |
Feng | Investigation of training data issues in ensemble classification based on margin concept: application to land cover mapping | |
EP4099228A1 (fr) | Apprentissage automatique sans annotation ameliore par regroupements adaptatifs en ensemble ouvert de classes | |
Santosa et al. | A robust feature construction for fish classification using grey Wolf optimizer | |
Karmakar et al. | Multilevel Random Forest algorithm in image recognition for various scientific applications | |
CN115293639A (zh) | 一种基于隐马尔可夫模型的战场态势研判方法 | |
CN115410250A (zh) | 阵列式人脸美丽预测方法、设备及存储介质 | |
EP4012619A1 (fr) | Méthode de compression d'un réseau de neurones artificiel | |
FR3114180A1 (fr) | Système et procédé pour éviter un oubli catastrophique dans un réseau neuronal artificiel | |
Mokeev | On application of convolutional neural network for classification of plant images | |
Ayala-Niño et al. | Evaluation of machine learning models to identify peach varieties based on leaf color | |
Altabaji et al. | Identification of Banana Leaf Diseases and Detection | |
EP3552155A1 (fr) | Procede et dispositif d'obtention d'un systeme de labellisation d'images | |
US20230290125A1 (en) | Image processing apparatus and image processing method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: UNKNOWN |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE |
|
PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |
|
17P | Request for examination filed |
Effective date: 20210921 |
|
AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |
|
DAV | Request for validation of the european patent (deleted) | ||
DAX | Request for extension of the european patent (deleted) | ||
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: EXAMINATION IS IN PROGRESS |
|
17Q | First examination report despatched |
Effective date: 20230731 |