An RGB-D semantic segmentation method based on cross-modal alignment fusion
By using the cross-modal alignment and fusion semantic segmentation network CCFN, and leveraging spatial channel attention and semantic flow correction modules, the problem of incomplete learning of feature relationships in RGB-D semantic segmentation is solved, achieving higher-precision object segmentation results.
CN117218348BActive Publication Date: 2026-06-16MINJIANG UNIVERSITY
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- MINJIANG UNIVERSITY
- Filing Date
- 2023-09-15
- Publication Date
- 2026-06-16
Smart Images
Figure CN117218348B_ABST
Abstract
The application relates to an RGB-D semantic segmentation method based on cross-modal alignment fusion. A spatial channel attention mechanism is used to fully utilize the spatial and channel relationship, so that both branches can pay attention to the complementary information of the other branch to solve the noise problem caused by the introduction of multi-modal, and a semantic flow correction module is introduced to effectively solve the problem that corresponding pixels between different modalities cannot be well aligned. Finally, a decoder with a semantic flow correction module is also used in the decoder part to make the deep semantic information more effectively transmitted to the shallow layer. Experiments verify the reliability and feasibility of the algorithm.
Need to check novelty before this filing date? Find Prior Art