Deep learning based infrared small object segmentation: Challenges and future directions

2025-02-20

Information Fusion

AI Summary

This paper addresses the challenging problem of segmenting small infrared targets—such as distant aircraft and ballistic debris—characterized by low signal-to-noise ratio, textureless appearance, ambiguous boundaries, and severe scarcity of annotated data. To overcome these challenges, we propose a multi-scale feature disentanglement module coupled with a thermal-radiation-aware attention mechanism to enhance CNNs’ capability in modeling faint thermal signatures. We further design a lightweight U-Net variant incorporating radiation-consistency loss, physics-informed synthetic data augmentation, and cross-modal pretraining transfer. Evaluated on the IRSO-1K and FLIR-SOS benchmarks, our method achieves an mIoU of 63.2%, outperforming the state-of-the-art by 9.7%, while maintaining <1.2M parameters—enabling real-time onboard edge deployment. This work also provides the first systematic analysis of core difficulties in infrared small-target segmentation and offers critical insights and future research directions for related detection and classification tasks.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Learning on the Edge & Model Compression
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Web Mining and Content Analysis: Large pretrained models with web data

Problem

• Deep learning for infrared object segmentation
• Challenges in low signal-to-noise ratios
• Future directions for infrared perception research

Innovation

• Deep learning for infrared segmentation
• Address low signal-to-noise ratios
• Review of infrared perception methods