MAPLE: Multi-Path Adaptive Propagation with Level-Aware Embeddings for Hierarchical Multi-Label Image Classification

arXiv cs.CV / 4/1/2026

📰 NewsSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

MAPLE is introduced as a framework for hierarchical multi-label classification in remote sensing, targeting cases where a single image may activate multiple taxonomic branches.
It combines level-aware embeddings initialized from graph-aware textual descriptions, graph structure encoding using GCNs, and an adaptive multi-modal fusion mechanism that balances semantic priors with visual evidence.
MAPLE uses an adaptive level-aware objective that automatically selects appropriate losses for each hierarchy level, improving how hierarchical dependencies are learned.
Experiments on CORINE-aligned Earth observation datasets (AID, DFC-15, MLRSNet) report consistent gains of up to +42% in few-shot settings with only ~2.6% parameter overhead.

Abstract

Hierarchical multi-label classification (HMLC) is essential for modeling structured label dependencies in remote sensing. Yet existing approaches struggle in multi-path settings, where images may activate multiple taxonomic branches, leading to underuse of hierarchical information. We propose MAPLE (Multi-Path Adaptive Propagation with Level-Aware Embeddings), a framework that integrates (i) hierarchical semantic initialization from graph-aware textual descriptions, (ii) graph-based structure encoding via graph convolutional networks (GCNs), and (iii) adaptive multi-modal fusion that dynamically balances semantic priors and visual evidence. An adaptive level-aware objective automatically selects appropriate losses per hierarchy level. Evaluations on CORINE-aligned remote sensing datasets (AID, DFC-15, and MLRSNet) show consistent improvements of up to +42% in few-shot regimes while adding only 2.6% parameter overhead, demonstrating that MAPLE effectively and efficiently models hierarchical semantics for Earth observation (EO).