Traitement en cours

Veuillez attendre...

Paramétrages

Paramétrages

Aller à Demande

1. CN106096670 - Cascaded convolutional neural network training method, device and system and cascaded convolutional neural network based image detection method, device and system

Office Chine
Numéro de la demande 201610439342.2
Date de la demande 17.06.2016
Numéro de publication 106096670
Date de publication 09.11.2016
Numéro de délivrance 106096670
Date de délivrance 30.07.2019
Type de publication B
CIB
G06K 9/66
GPHYSIQUE
06CALCUL; COMPTAGE
KRECONNAISSANCE DES DONNÉES; PRÉSENTATION DES DONNÉES; SUPPORTS D'ENREGISTREMENT; MANIPULATION DES SUPPORTS D'ENREGISTREMENT
9Méthodes ou dispositions pour la lecture ou la reconnaissance de caractères imprimés ou écrits ou pour la reconnaissance de formes, p.ex. d'empreintes digitales
62Méthodes ou dispositions pour la reconnaissance utilisant des moyens électroniques
64utilisant des comparaisons ou corrélations simultanées de signaux images avec une pluralité de références, p.ex. matrice de résistances
66avec des références réglables par une méthode adaptative, p.ex. en s'instruisant
G06N 3/08
GPHYSIQUE
06CALCUL; COMPTAGE
NSYSTÈMES DE CALCULATEURS BASÉS SUR DES MODÈLES DE CALCUL SPÉCIFIQUES
3Systèmes de calculateurs basés sur des modèles biologiques
02utilisant des modèles de réseaux neuronaux
08Méthodes d'apprentissage
G06T 7/00
GPHYSIQUE
06CALCUL; COMPTAGE
TTRAITEMENT OU GÉNÉRATION DE DONNÉES D'IMAGE, EN GÉNÉRAL
7Analyse d'image
CPC
G06K 9/66
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
9Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
62Methods or arrangements for recognition using electronic means
64using simultaneous comparisons or correlations of the image signals with a plurality of references, e.g. resistor matrix
66references adjustable by an adaptive method, e.g. learning
G06N 3/084
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
08Learning methods
084Back-propagation
G06N 3/086
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
08Learning methods
086using evolutionary programming, e.g. genetic algorithms
G06T 2207/20081
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
2207Indexing scheme for image analysis or image enhancement
20Special algorithmic details
20081Training; Learning
G06T 2207/20084
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
2207Indexing scheme for image analysis or image enhancement
20Special algorithmic details
20084Artificial neural networks [ANN]
Déposants BEIJING SENSETIME SCIENCE & TECHNOLOGY DEVELOPMENT CO., LTD.
深圳市商汤科技有限公司
Inventeurs QIN HONGWEI
秦红伟
YAN DUNJIE
闫俊杰
Mandataires 北京三聚阳光知识产权代理有限公司 11250
Titre
(EN) Cascaded convolutional neural network training method, device and system and cascaded convolutional neural network based image detection method, device and system
(ZH) 级联卷积神经网络训练和图像检测方法、装置及系统
Abrégé
(EN)
The invention discloses a cascaded convolutional neural network training method, device and system and a cascaded convolutional neural network based image detection method, device and system. The training method comprises the steps of processing image data of at least a local region of an image to be learnt into image data of N types of input regions which are different in size, wherein N is an integer which is greater than or equal to 2; taking the image data of the N types of input regions as input of each-grade convolutional neural network in the N-grade cascaded convolutional neural network respectively, and carrying out training on each-grade convolutional neural network; and correlating at least one training result outputted by each-grade convolutional neural network respectively, transmitting the correlated training results back to each-grade convolutional neural network so as to adjust parameters of each-grade neural network. According to the invention, the parameters of each-grade neural network can be adjusted when the training results are broadcast to each-grade convolutional neural network, so that the cascaded convolutional neural network is enabled to achieve global optimization of the neural network parameters in training.

(ZH)
本发明公开一种级联卷积神经网络训练和图像检测方法、装置及系统,其中,所述训练方法包括:将待学习图像至少局部区域的图像数据分别处理成N种不同大小的输入区域的图像数据,N为大于或等于2的整数;分别将N种输入区域的图像数据作为N级级联的卷积神经网络中各级卷积神经网络的输入,对各级卷积神经网络进行训练;将各级卷积神经网络分别输出的至少一训练结果进行关联,并将关联后的训练结果回传至各级卷积神经网络以调整各级神经网络的参数。在将训练结果传播至各级卷积神经网络时,能够调整各级神经网络的参数,使得级联卷积神经网络在训练时能够达到神经网络参数的全局优化。