A semantic segmentation method and device integrating optical and SAR channel correlation
By extracting the feature maps of optical images and SAR images and fusion of channel attention mechanisms, the problem of lack of semantic information in the fusion of optical images and SAR images in the prior art is solved, and high-quality semantic segmentation of high-resolution remote sensing images is achieved.
CN118115732BActive Publication Date: 2025-04-04AEROSPACE INFORMATION RES INST CAS
6 Cites 0 Cited by
Patent Information
- Application Number
- CN202410052659.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-12
- Publication Date
- 2025-04-04
- Estimated Expiration
- 2044-01-12
AI Technical Summary
Technical Problem
In the prior art, optical image and SAR image fusion method lacks deep fusion of semantic information, resulting in low quality of the result.
Method used
A semantic segmentation method that fuses the correlation between optical and SAR channels is proposed. By extracting the feature maps of optical images and SAR images, the channel attention mechanism is fused, and the encoder-decoder architecture and jump connection are upsampled layer by layer to improve segmentation accuracy.
Benefits of technology
The semantic segmentation of high-resolution remote sensing images is realized, the complementarity of heterogeneous data is effectively utilized, and the quality of the fusion result between optical images and SAR images is improved.
✦ Generated by Eureka AI based on patent content.
Abstract
The present application discloses a semantic segmentation method and device that fuses the correlation between optical and SAR channels, which solves the problem that the existing optical and SAR fusion methods based on deep learning lack deep fusion of semantic information. The semantic segmentation method that fuses the correlation between optical and SAR channels includes the following steps: acquiring optical images and SAR images of the study area; extracting feature maps of the optical image and the SAR image respectively by extracting features; the feature maps of the two images contain corresponding input feature maps at multiple levels; fusing the feature maps of the optical image and the SAR image at the same level to obtain a single-level result feature map; adding all the single-level result feature maps layer by layer from deep to shallow and convolution upsampling to obtain the result image. The present application realizes the semantic segmentation of high-resolution remote sensing images, which improves the overall segmentation accuracy of the network.
Need to check novelty before this filing date? Find Prior Art