MACL: Multi-Label Adaptive Contrastive Learning Loss for Remote Sensing Image Retrieval

2025-12-18

AI Summary

To address semantic overlap, long-tailed label distribution, and complex label co-occurrence in remote sensing image multi-label retrieval, this paper proposes a Multi-Label Adaptive Contrastive Learning (ML-ACL) framework. Methodologically, it introduces three key components: (1) a label-aware hard negative sampling strategy to mitigate semantic confusion; (2) a frequency-sensitive weighting mechanism that explicitly models label distribution skewness; and (3) dynamic temperature scaling to adaptively adjust inter-class discriminative granularity. Evaluated on three major benchmarks—DLRSD, ML-AID, and WHDLD—the framework consistently outperforms existing contrastive learning methods, achieving average improvements of 3.2–5.8 percentage points in mAP and Recall@K. It effectively alleviates semantic imbalance, enhances representation learning for rare land-cover classes, and improves cross-scene retrieval robustness.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Computer Vision: Image and Video Retrieval
• Multiagent Systems: Multiagent Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Semantic overlap among land-cover categories, highly imbalanced label distributions, and complex inter-class co-occurrence patterns constitute significant challenges for multi-label remote-sensing image retrieval. In this article, Multi-Label Adaptive Contrastive Learning (MACL) is introduced as an extension of contrastive learning to address them. It integrates label-aware sampling, frequency-sensitive weighting, and dynamic-temperature scaling to achieve balanced representation learning across both common and rare categories. Extensive experiments on three benchmark datasets (DLRSD, ML-AID, and WHDLD), show that MACL consistently outperforms contrastive-loss based baselines, effectively mitigating semantic imbalance and delivering more reliable retrieval performance in large-scale remote-sensing archives. Code, pretrained models, and evaluation scripts will be released at https://github.com/amna/MACL upon acceptance.

Problem

• Addresses semantic overlap in multi-label remote sensing image retrieval
• Mitigates imbalanced label distributions across land-cover categories
• Handles complex inter-class co-occurrence patterns in large archives

Innovation

• Adaptive contrastive learning for multi-label retrieval
• Label-aware sampling and frequency-sensitive weighting
• Dynamic-temperature scaling for balanced representation learning