LiTANet: Lightweight Semantic Segmentation Network with Triple Attention and Depthwise Separable Convolution for Remote Sensing Imagery

This paper presents LiTANet, a lightweight semantic segmentation network designed for high-resolution remote sensing imagery. LiTANet integrates depthwise separable convolution with a triple attention mechanism to enhance feature representation while significantly reducing computational cost. A robust encoder–decoder architecture with H-Swish activation further strengthens non-linear fitting ability and improves feature extraction efficiency. Extensive experiments conducted on the ISPRS Potsdam and Vaihingen datasets demonstrate that LiTANet achieves competitive segmentation accuracy, with Avg.F1 scores of 89.2% and 88.4%, respectively, while using only 10.5M parameters, substantially fewer than traditional deep models. Additional ablation experiments verify the effectiveness of each key module. The results show that LiTANet achieves an excellent balance between accuracy, generalization capability, and model efficiency, providing a new option for resource-constrained remote sensing applications.

• Publication year

  2025

• Venue

  2025 2nd International Conference on Image, Signal Processing and Communication Technology (ISPCT)

• Publication date

  2025-12-05

• Fields of study

  Not labeled

• Identifiers
  DOI 10.1109/ISPCT68220.2025.11406769
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Early Wildfire Detection and Distance Estimation Using Aerial Visible-Infrared Images

  2024cited by this paper
• Multispectral Semantic Segmentation for Land Cover Classification: An Overview

  2024cited by this paper
• MQANet: Multi-Task Quadruple Attention Network of Multi-Object Semantic Segmentation from Remote Sensing Images

  2022cited by this paper
• Searching for MobileNetV3

  2019cited by this paper
• Xception: Deep Learning with Depthwise Separable Convolutions

  2016cited by this paper
• Neural Machine Translation by Jointly Learning to Align and Translate

  2014cited by this paper

• No citing papers are available for this paper.