文献类型：期刊文献

英文题名：TM-KAN:A Dual-Branch Encoder with Transformer and Mamba for Remote Sensing Image Semantic Segmentation using KAN-based Decoder

作者：Zhang, Memgmeng[1]; Ding, Bo[1]; Jing, Hongyuan[1]; Liu, Yuting[1]

第一作者：Zhang, Memgmeng

机构：[1] Beijing Key Laboratory of Information Service Engineering, Beijing Union University, Beijing, 100101, China

第一机构：北京联合大学北京市信息服务工程重点实验室

年份：2025

外文期刊名：TechRxiv

收录：EI(收录号：20250314504)

语种：英文

外文关键词：Complex networks - Computational complexity - Decoding - Deep learning - Disaster prevention - Environmental management - Environmental technology - Extraction - Information management - Semantic Segmentation - Semantic Web - Semantics - Signal encoding - Space optics - Urban planning

摘要：Remote sensing technology plays a pivotal role in environmental monitoring, urban planning, disaster assessment, and resource management, with semantic segmentation being critical for extracting meaningful information from images. Traditional methods relying on manual feature extraction have been surpassed by deep learning, particularly Transformer architectures, which excel in modeling long-range dependencies and global context. However, their quadratic computational complexity and limited ability to capture fine-grained local features pose challenges for high-resolution remote sensing image segmentation. To address these limitations, this paper proposes a novel dual-branch semantic segmentation framework that integrates Transformer and Mamba architectures with a Kolmogorov-Arnold Network (KAN)-based decoder. The Transformer branch captures global context, while the Mamba branch, leveraging its linear-complexity selective state space mechanism, efficiently handles long sequences and multi-scale features. A feature fusion mechanism combines these representations, and the KAN decoder employs learnable univariate functions to model complex nonlinear relationships, enhancing boundary delineation and feature reconstruction. Key contributions include the dual-branch encoder design, an effective fusion strategy, and the innovative KAN decoder. Extensive experiments on datasets like WHU-Building and Massachusetts Roads demonstrate promising segmentation performance, offering a significant advancement in balancing efficiency and accuracy for remote sensing applications. ? 2025, CC BY.