Title:
METHOD AND DEVICE FOR OPTIMIZING NEURAL NETWORK-BASED TARGET CLASSIFICATION MODEL
Document Type and Number:
WIPO Patent Application WO/2021/135607
Kind Code:
A1
Abstract:
A method and device for optimizing a neural network-based target classification model. The method comprises: constructing a neural network-based target classification model, training the target classification model, and using the trained target classification model to classify a target image (S110); and when a new target image appears, and the new target image corresponds to a new target situation, and can still be classified according to an original classification system, determining a recognition result of classification performed by the target classification model on the new target image, and if the target classification model cannot correctly classify the new target image, selecting a portion of parameters according to the new target image, adjusting the portion of parameters, and acquiring, by means of training, a target classification model that can correctly classify the new target image (S120). The above solution ensures the performance of the target classification model, and enables the target classification model to adapt to new scenarios. In addition, training speed is greatly improved, and model training can be completed within about 20 minutes.
Inventors:
DI SHUNRAN (CN)
ZHANG YIFAN (CN)
LIU JIE (CN)
TIAN JIFENG (CN)
ZHANG YIFAN (CN)
LIU JIE (CN)
TIAN JIFENG (CN)
Application Number:
PCT/CN2020/125346
Publication Date:
July 08, 2021
Filing Date:
October 30, 2020
Export Citation:
Assignee:
GOERTEK INC (CN)
International Classes:
G06K9/62
Foreign References:
| CN111178446A | 2020-05-19 | |||
| CN110321964A | 2019-10-11 | |||
| CN110211123A | 2019-09-06 | |||
| CN110210560A | 2019-09-06 | |||
| US20190385059A1 | 2019-12-19 | |||
| CN110472681A | 2019-11-19 |
Download PDF: