STAR-IOD: Scale-decoupled Topology Alignment with Pseudo-label Refinement for Remote Sensing Incremental Object Detection

2026-05-20

AI Summary

This work addresses the challenges of incremental object detection in remote sensing imagery, where scale variations induce intra-class representation inconsistency and missing annotations for old classes lead to catastrophic forgetting. To tackle these issues, the authors propose the STAR-IOD framework, which explicitly models scale-disentangled topological alignment for the first time in remote sensing scenarios. It employs Subspace Topological Distillation (STD) to preserve inter-class structural consistency and introduces a plug-in Clustering-driven Pseudo-label Generator (CPG) that dynamically sets class-specific thresholds via K-Means clustering to mitigate the impact of missing labels. Evaluated on DIOR-IOD and DOTA-IOD benchmarks, the method improves mAP by 1.7% and 2.1%, respectively, effectively alleviating catastrophic forgetting while maintaining strong detection performance for both base and novel classes.

Abstract

Remote sensing imagery typically arrives in the form of continuous data streams. Traditional detectors often forget previously learned categories when learning new ones; therefore, research on Remote Sensing Incremental Object Detection (RS-IOD) is of great significance. However, existing methods largely overlook the intra-class scale variations prevalent in remote sensing scenes, which undermines the effectiveness of knowledge transfer and old knowledge preservation. Moreover, RS-IOD also suffers from missing annotations, which cause the model to misclassify old-class instances as background. To address these challenges, we propose a novel framework, STAR-IOD. First, we introduce a Subspace-decoupled Topology Distillation (STD) module to transfer structural knowledge, explicitly aligning inter-class topological relationships and mitigating intra-class representation discrepancies induced by scale shifts. Furthermore, we introduce the Clustering-driven Pseudo-label Generator (CPG), a plug-and-play module that leverages K-Means clustering to dynamically identify class-specific thresholds, thereby guaranteeing an accurate distinction between true positive targets and background noise and alleviating the issue of missing annotations for old classes. We also constructed two Remote Sensing Incremental Object Detection datasets, DIOR-IOD and DOTA-IOD to facilitate research on RS-IOD. Extensive experiments demonstrate that our method outperforms state-of-the-art approaches by 1.7% and 2.1% mAP on DIOR-IOD and DOTA-IOD, respectively, effectively alleviating catastrophic forgetting while preserving strong detection performance on both base and novel classes. The code and dataset are released at: https://github.com/zyt95579/STAR-IOD.

Problem

• Remote Sensing Incremental Object Detection
• intra-class scale variation
• missing annotations
• catastrophic forgetting
• knowledge preservation

Innovation

• Incremental Object Detection
• Scale Variation
• Topology Distillation
• Pseudo-label Refinement
• Remote Sensing