Boosting Infrared Small Target Detection via Logit-Domain Contrast and Adaptive Shape Refinement

2026-07-02 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

A new method, Adaptive-Contrastive SLSIoU (AC-SLSIoU), is proposed to enhance Infrared Small Target Detection (IRSTD), addressing challenges like tiny target size, low signal-to-noise ratio, and blurred boundaries. Existing IRSTD approaches often struggle with weak-target discrimination due to saturated probabilities in post-activation supervision and produce halo-like predictions from thermal diffusion. AC-SLSIoU integrates three key components: a Logit-Domain Margin Constraint (LDMC) to increase the response gap between targets and hard negatives in the logit space, Adaptive Boundary Suppression (ABS) for scale-aware annular penalties to refine target contours, and False-Alarm Focal Loss to penalize high-confidence false alarms. This plug-and-play loss function requires no extra inference overhead, seamlessly integrates into current detectors, and consistently improves both detection accuracy and shape quality, as demonstrated by extensive experiments and cross-backbone evaluations.

Key takeaway

For Computer Vision Engineers developing Infrared Small Target Detection (IRSTD) systems, consider integrating Adaptive-Contrastive SLSIoU (AC-SLSIoU) into your existing detectors. This plug-and-play loss function, which includes logit-domain contrast and adaptive boundary suppression, directly addresses weak-target discrimination and blurred boundary issues. Implementing AC-SLSIoU can significantly improve both detection accuracy and the quality of target shapes without adding inference overhead, streamlining your model's performance in complex scenes.

Key insights

Enhancing IRSTD requires logit-domain contrast, adaptive boundary refinement, and false-alarm focused loss to overcome weak-target and blurred boundary issues.

Principles

• Logit-domain contrast improves weak-target discrimination.
• Scale-aware annular penalties refine target contours.
• Weighting high-probability negatives reduces false alarms.

Method

AC-SLSIoU combines Logit-Domain Margin Constraint, Adaptive Boundary Suppression, and False-Alarm Focal Loss to create a plug-and-play, discriminative, and shape-aware loss function for IRSTD.

In practice

• Integrate AC-SLSIoU into existing detectors.
• Apply scale-aware penalties for contour refinement.
• Use focal loss to target high-confidence false alarms.

Topics

• Infrared Small Target Detection
• Computer Vision
• Loss Functions
• Logit-Domain Contrast
• Adaptive Boundary Suppression
• False-Alarm Focal Loss

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• Selective Attention-Based Network for Robust Infrared Small Target Detection — SANet enhances infrared small target detection by dynamically fusing multi-scale features and refining early feature extraction.
• HAMSA: Scanning-Free Vision State Space Models via SpectralPulseNet — HAMSA introduces a scanning-free Vision SSM operating in the spectral domain for improved efficiency and performance.
• From Spatial to Spectral: An Efficient, Frequency-Guided Feature Representation Learner for Small Object Detection — Shifting small object detection from spatial to spectral processing improves efficiency and accuracy by preserving high-frequency features.
• Frequency-Enhanced Dual-Subspace Networks for Few-Shot Fine-Grained Image Classification — FEDSNet improves few-shot fine-grained image classification by integrating frequency domain features to enhance structural stability and reduce overfitting.
• Real-Time Source-Free Object Detection — RT-SFOD achieves state-of-the-art source-free object detection by optimizing pseudo-label fusion and multi-scale feature diversification for speed and accuracy.
• ATT-CR: Adaptive Triangular Transformer for Cloud Removal — Adaptive Triangular Transformer (ATT-CR) efficiently removes clouds from remote sensing images by combining linear triangular attention and feature-selective gating.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.