A dual-branch transformer-mamba network integrating cross-scale wavelet-mamba fusion for semantic segmentation of grapevine stems and branches
Smart Agricultural Technology · 2026-12 · 15 · 102651
DOI: 10.1016/j.atech.2026.102651 ↗Kurz gefasst
Fachartikel in der Zeitschrift Smart Agricultural Technology, 2026-12. Verfasst von Long Song und 5 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Accurate identification of grapevine stems and branches is crucial for automated pruning and vineyard management, yet the task remains difficult due to thin structures, occlusions, and complex outdoor backgrounds. To address these difficulties, we propose a Dual-Branch Transformer-Mamba Fusion Network (DTMF) that integrates global context, frequency–spatial decoupling, and boundary-aware optimization. The framework consists of a ViT-based encoder for long-range context modelling and a DWT-Mamba encoder that exploits multi-frequency features with linear-complexity state-space modelling. A cross-scale Wavelet-Mamba fusion module and dual-decoder design preserve complementary feature advantages, while training uses a two-stage composite loss. The reported internal comparison gives 94.24% mPA and 88.25% mIoU on the grapevine pruning dataset. External images provide a qualitative illustration rather than quantitative verification of cross-domain generalization. This frequency-spatial dual-branch design provides a candidate approach to fine-grained structure recognition and offers a potential perception input for future automated pruning systems.
Zitieren
APA
Song, L., Song, Z., He, L., Lyu, X., Bao, P., & Shen, C. (2026). A dual-branch transformer-mamba network integrating cross-scale wavelet-mamba fusion for semantic segmentation of grapevine stems and branches. Smart Agricultural Technology, 15, 102651. https://doi.org/10.1016/j.atech.2026.102651
Vancouver
Song L, Song Z, He L, Lyu X, Bao P, Shen C. A dual-branch transformer-mamba network integrating cross-scale wavelet-mamba fusion for semantic segmentation of grapevine stems and branches. Smart Agricultural Technology. 2026;15:102651. doi:10.1016/j.atech.2026.102651
BibTeX
@article{Song2026,
author = {Song, Long and Song, Zhenshuai and He, Lei and Lyu, Xuefeng and Bao, Pengyu and Shen, Congju},
title = {A dual-branch transformer-mamba network integrating cross-scale wavelet-mamba fusion for semantic segmentation of grapevine stems and branches},
journal = {Smart Agricultural Technology},
year = {2026},
volume = {15},
pages = {102651},
doi = {10.1016/j.atech.2026.102651}
}
EUVIMED