Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PPMamba: A Pyramid Pooling Local Auxiliary SSM-Based Model for Remote Sensing Image Semantic Segmentation

Yin Hu, Xianping Ma, Jialu Sui, Man-On Pun

TL;DR

The paper addresses the challenge of semantic segmentation in remote sensing by balancing local detail with long-range context under linear computational complexity. It introduces PPMamba, a CNN-Mamba hybrid that leverages PP-SSM blocks and an omnidirectional OSS to fuse multiscale local features with global dependencies in a UNet-like encoder–decoder. The core contributions are the multi-branch pyramid pooling within PP-SSM and the eight-direction OSS, which together preserve local semantics while modeling long-range structure. Empirical results on ISPRS Vaihingen and LoveDA Urban demonstrate competitive or superior performance compared with state-of-the-art baselines, with ablations confirming the effectiveness of the multi-branch and pyramid pooling design and an emphasis on computational efficiency for practical RS applications.

Abstract

Semantic segmentation is a vital task in the field of remote sensing (RS). However, conventional convolutional neural network (CNN) and transformer-based models face limitations in capturing long-range dependencies or are often computationally intensive. Recently, an advanced state space model (SSM), namely Mamba, was introduced, offering linear computational complexity while effectively establishing long-distance dependencies. Despite their advantages, Mamba-based methods encounter challenges in preserving local semantic information. To cope with these challenges, this paper proposes a novel network called Pyramid Pooling Mamba (PPMamba), which integrates CNN and Mamba for RS semantic segmentation tasks. The core structure of PPMamba, the Pyramid Pooling-State Space Model (PP-SSM) block, combines a local auxiliary mechanism with an omnidirectional state space model (OSS) that selectively scans feature maps from eight directions, capturing comprehensive feature information. Additionally, the auxiliary mechanism includes pyramid-shaped convolutional branches designed to extract features at multiple scales. Extensive experiments on two widely-used datasets, ISPRS Vaihingen and LoveDA Urban, demonstrate that PPMamba achieves competitive performance compared to state-of-the-art models.

PPMamba: A Pyramid Pooling Local Auxiliary SSM-Based Model for Remote Sensing Image Semantic Segmentation

TL;DR

The paper addresses the challenge of semantic segmentation in remote sensing by balancing local detail with long-range context under linear computational complexity. It introduces PPMamba, a CNN-Mamba hybrid that leverages PP-SSM blocks and an omnidirectional OSS to fuse multiscale local features with global dependencies in a UNet-like encoder–decoder. The core contributions are the multi-branch pyramid pooling within PP-SSM and the eight-direction OSS, which together preserve local semantics while modeling long-range structure. Empirical results on ISPRS Vaihingen and LoveDA Urban demonstrate competitive or superior performance compared with state-of-the-art baselines, with ablations confirming the effectiveness of the multi-branch and pyramid pooling design and an emphasis on computational efficiency for practical RS applications.

Abstract

Semantic segmentation is a vital task in the field of remote sensing (RS). However, conventional convolutional neural network (CNN) and transformer-based models face limitations in capturing long-range dependencies or are often computationally intensive. Recently, an advanced state space model (SSM), namely Mamba, was introduced, offering linear computational complexity while effectively establishing long-distance dependencies. Despite their advantages, Mamba-based methods encounter challenges in preserving local semantic information. To cope with these challenges, this paper proposes a novel network called Pyramid Pooling Mamba (PPMamba), which integrates CNN and Mamba for RS semantic segmentation tasks. The core structure of PPMamba, the Pyramid Pooling-State Space Model (PP-SSM) block, combines a local auxiliary mechanism with an omnidirectional state space model (OSS) that selectively scans feature maps from eight directions, capturing comprehensive feature information. Additionally, the auxiliary mechanism includes pyramid-shaped convolutional branches designed to extract features at multiple scales. Extensive experiments on two widely-used datasets, ISPRS Vaihingen and LoveDA Urban, demonstrate that PPMamba achieves competitive performance compared to state-of-the-art models.
Paper Structure (22 sections, 7 equations, 7 figures, 4 tables)

This paper contains 22 sections, 7 equations, 7 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Remote Sensing Image Semantic Segmentation
  4. Mamba
  5. Spatial Pyramid Pooling
  6. Methodology
  7. Proposed PPMamba
  8. Proposed PP-SSM Block
  9. Omnidirectional State Space Block (OSS)
  10. Experiments
  11. Datasets
  12. ISPRS Vaihingen
  13. LoveDA Urban
  14. Evaluation Metrics
  15. Implementation Details
  16. ...and 7 more sections

Figures (7)

  • Figure 1: The architecture of proposed PPMamba.
  • Figure 2: The architectures of the conventional visual SSM block and the proposed PP-SSM block. (a) The architecture of a conventional visual SSM block. (b) Proposed architecture of PP-SSM Block.
  • Figure 3: (a) The architecture of the proposed OSS block. (b) The illustration of the selective scan directions of OSSM.
  • Figure 4: Illustration of the operation of the proposed oriented scanning module (OSSM).
  • Figure 5: Performance comparisons on the ISPRS Vaihaigen dataset with the size of $1024 \times 1024$. (a) NIRRG images, (b) Ground truth, (c) ABCNet, (d) MANet, (e) CMTFNet, (f) UNetFormer, (g) FTUNetFormer, (h) HST_UNet, (i) TransUNet, (j) RS3Mamba, (k) RS-Mamba and (l) PPMamba.
  • ...and 2 more figures