PulseTrendingJournal ClubResearchersJournalsExplore
Instagram
HomeTrendingJournal ClubExplore
Synapse
⌘+K
Synapse
June 3, 2026Remote SensingOpen Access

MAFT: A Lightweight Network for Martian Rock Segmentation Based on an Adaptive Frequency Transformer

View Full Paper
Ask AI
Bookmark
Share

Authors

CLChu LiYJYutong JiaGWGang Wan

Discussion

Loading...

Member takes

Overview

Randomized trial demonstrates enhanced segmentation accuracy in Martian rocks, suggesting improved rover navigation capabilities.

Key Points

  • This research aims to develop a lightweight network for accurate segmentation of Martian rocks, essential for rover navigation.
  • Proposed MAFT, building on Adaptive Frequency Transformer with improved backbone (IAFFormer).
  • Replaced traditional self-attention with frequency-domain approach, reducing complexity from quadratic to linear.
  • Developed dataset of Martian rocks for Zhurong rover and conducted experiments on synthetic and real datasets.
  • MAFT achieved pixel accuracy of 98.17% and IoU of 88.90% on TWMARS-V2 dataset.
  • Achieved highest segmentation accuracy with only 2.97 M parameters and 15.49 G FLOPs.
  • Enhanced module design improves boundary perception without increasing parameter count.

Cite This Study

Li et al. (2026) studied this question.

synapsesocial.com/papers/6a1fc509dee9eb8c0dce67dehttps://doi.org/10.3390/rs18111794
View Full Paper
Ask AI
Bookmark
Share