Title:
NON-ANCHOR MECHANISM-BASED FLAME DETECTION METHOD AND APPARATUS
Document Type and Number:
WIPO Patent Application WO/2021/114221
Kind Code:
A1
Abstract:
Provided in the present application are a non-anchor mechanism-based flame detection method and apparatus. The method comprises: obtaining a flame detection data set; constructing a RetinaNet network model having a non-anchor mechanism-based feature selection module; for each image instance, when the image instance is mapped to a specific feature pyramid layer, defining an effective area, ineffective area, and neglected area between the two according to a scale factor; by means of classifying sub-networks and regression sub-networks, predicting the probability of each anchor point belonging to a different category and the offset of each anchor point box from the nearest instance; and selecting, as the learning level of the instance, the feature layer that has the smallest sum of classification loss of the classifying sub-network and regression loss of the regression sub-network, and training the RetinaNet network model. The present application can complete the entire detection process from original image input to flame position information output by means of a deep learning algorithm, and to a certain extent prevents situations of misreporting and underreporting that are present in manual design features.
Inventors:
WANG XINXIN (CN)
YUN ZHOUHUI (CN)
YE CHAO (CN)
WU BIN (CN)
XIE JIPENG (CN)
YING YANLI (CN)
HUANG JIANGLIN (CN)
WANG XU (CN)
JIA NAN (CN)
LAI ZEWEI (CN)
YUN ZHOUHUI (CN)
YE CHAO (CN)
WU BIN (CN)
XIE JIPENG (CN)
YING YANLI (CN)
HUANG JIANGLIN (CN)
WANG XU (CN)
JIA NAN (CN)
LAI ZEWEI (CN)
Application Number:
PCT/CN2019/125110
Publication Date:
June 17, 2021
Filing Date:
December 13, 2019
Export Citation:
Assignee:
JIANGXI HONGDU AVIATION IND GROUP COMPANY LTD (CN)
International Classes:
G06K9/00
Foreign References:
| CN110348390A | 2019-10-18 | |||
| CN108647559A | 2018-10-12 |
Other References:
ZHU CHENCHEN; HE YIHUI; SAVVIDES MARIOS: "Feature Selective Anchor-Free Module for Single-Shot Object Detection", 2019 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), IEEE, 15 June 2019 (2019-06-15), pages 840 - 849, XP033686623, DOI: 10.1109/CVPR.2019.00093
KIM BYOUNGJUN, LEE JOONWHOAN: "A Video-Based Fire Detection Using Deep Learning Models", APPLIED SCIENCES, vol. 9, no. 14, 1 January 2019 (2019-01-01), pages 2862, XP055820803, DOI: 10.3390/app9142862
KIM BYOUNGJUN, LEE JOONWHOAN: "A Video-Based Fire Detection Using Deep Learning Models", APPLIED SCIENCES, vol. 9, no. 14, 1 January 2019 (2019-01-01), pages 2862, XP055820803, DOI: 10.3390/app9142862
Attorney, Agent or Firm:
BEIJING HANGXIN HIGH-TECH INTELLECTUAL PROPERTY AGENCY (CN)
Download PDF:
Previous Patent: FULL-AUTONOMOUS UNMANNED MOWING SYSTEM
Next Patent: METHOD FOR MANUFACTURING SEMICONDUCTOR STRUCTURE
Next Patent: METHOD FOR MANUFACTURING SEMICONDUCTOR STRUCTURE