Title:
DOUBLE-FEATURE FUSION SEMANTIC SEGMENTATION SYSTEM AND METHOD BASED ON INTERNET OF THINGS PERCEPTION
Document Type and Number:
WIPO Patent Application WO/2022/227913
Kind Code:
A1
Abstract:
The present application discloses a double-feature fusion semantic segmentation system and method based on Internet of Things perception. The method comprises the steps of: S1, performing feature encoding on an original image to obtain features of different scales; S2, learning the features of different scales by means of two attention refining blocks to obtain a multi-level fusion feature; S3, performing dimensionality reduction on the multi-level fusion feature to obtain a dimensionality-reduced feature; S4, performing context encoding on the dimensionality-reduced feature by using depthwise factorized convolution of different convolution scales to obtain local features of different scales; S5, performing global pooling on the dimensionality-reduced feature by using a global average pooling layer to obtain a global feature; S6, performing channel splicing fusion on the global feature and the local features to obtain a multi-scale context fusion feature; S7, performing channel splicing fusion on the dimensionality-reduced feature and the multi-scale context fusion feature to obtain a splicing feature; and S8, obtaining an output according to the splicing feature. The semantic difference among multi-level features is relieved, the information representation is enriched, and the recognition precision is improved.
Inventors:
ZHU XINZHONG (CN)
XU HUIYING (CN)
TU WENXUAN (CN)
ZHAO JIANMIN (CN)
XU HUIYING (CN)
TU WENXUAN (CN)
ZHAO JIANMIN (CN)
Application Number:
PCT/CN2022/081427
Publication Date:
November 03, 2022
Filing Date:
March 17, 2022
Export Citation:
Assignee:
UNIV ZHEJIANG NORMAL (CN)
International Classes:
G06K9/62
Foreign References:
| CN113221969A | 2021-08-06 | |||
| CN112651973A | 2021-04-13 | |||
| CN111210432A | 2020-05-29 | |||
| US20150104102A1 | 2015-04-16 |
Other References:
TANG XIANGYAN, TU WENXUAN, LI KEQIU, CHENG JIEREN: "DFFNet: An IoT-perceptive dual feature fusion network for general real-time semantic segmentation", INFORMATION SCIENCES, ELSEVIER, AMSTERDAM, NL, vol. 565, 1 July 2021 (2021-07-01), AMSTERDAM, NL, pages 326 - 343, XP055981882, ISSN: 0020-0255, DOI: 10.1016/j.ins.2021.02.004
Attorney, Agent or Firm:
BEIJING TIANDA INTELLECTUAL PROPERTY OFFICE (CN)
Download PDF: