LDFE: Laplacian Decoupled Feature Enhancement Block for Dual-Stream CNN-based RGB-IR Object Detection

2026-07-11 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

The Laplacian Decoupled Feature Enhancement (LDFE) block significantly improves RGB-IR object detection, particularly under extreme conditions, by effectively fusing complementary information from dual-stream CNN backbones. LDFE employs a four-step paradigm: Laplacian Pyramid-based global-local feature decomposition, denoising, fusion, and reconstruction. It utilizes a Global State Space Enhancement module (GS2E) for global features, incorporating cross-modal attention and a State Space Model, and a Local Convolutional Correlation Enhancement module (LC2E) for local features, applying L1 Normalization and multi-dimensional attention. This approach achieved mAP improvements of 6.2% on M3FD, 3.7% on DroneVehicle, 4.7% on LLVIP, 2.3% on FLIR-Aligned, 4.1% on KAIST, and 2.0% on VEDAI datasets, while maintaining high efficiency and low computational resources.

Key takeaway

For Machine Learning Engineers developing robust object detection systems for challenging RGB-IR environments, consider integrating the Laplacian Decoupled Feature Enhancement (LDFE) block. Its global-local feature decomposition and tailored denoising strategies significantly improve detection accuracy and efficiency. Implementing LDFE within your YOLO-based dual-stream CNNs can reduce missed detections and false positives, especially under low-light or adverse weather conditions.

Key insights

Decoupling global and local features with tailored denoising and fusion significantly boosts RGB-IR object detection performance.

Principles

• RGB-IR complementarity improves detection in extreme conditions.
• Decompose features into global and local components for specialized processing.
• Tailored denoising and fusion strategies enhance multimodal feature integration.

Method

LDFE employs Laplacian decomposition, followed by denoising and fusion via GS2E (global, State Space Model, alternating attention) and LC2E (local, L1 Normalization, multi-dimensional attention), then Laplacian reconstruction.

In practice

• Integrate LDFE into YOLO-based dual-stream CNNs.
• Apply Laplacian Pyramid for multi-scale feature decomposition.
• Utilize State Space Models for efficient global feature extraction.

Topics

• RGB-IR Object Detection
• Feature Fusion
• Laplacian Pyramid
• Dual-Stream CNN
• State Space Models
• YOLO

Code references

• ultralytics/ultralytics

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

Related on AIssential

• LDFE: Laplacian Decoupled Feature Enhancement Block for Dual-Stream CNN-based RGB-IR Object Detection — The LDFE block enhances RGB-IR object detection by decoupling features into global and local components for specialized fusion and denoising.
• 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.
• DUALVISION: RGB-Infrared Multimodal Large Language Models for Robust Visual Reasoning — Integrating infrared and RGB data via localized cross-attention significantly boosts MLLM robustness in degraded visual conditions.
• Focus, Align, and Sustain: Counteracting Gradient Dilution in Incremental Object Detection — Gradient Dilution destabilizes incremental object detection; FAS counteracts it by focusing, aligning, and sustaining gradient flow.
• MSRNet: A Multi-Scale Recursive Network for Camouflaged Object Detection — MSRNet combines multi-scale transformer features and recursive decoding to precisely detect camouflaged objects, including small and multiple instances.
• Dual-Integrated Low-Latency Single-Lens Infrared Computational Imaging for Object Detection — Integrating optical physics priors with deep learning for joint reconstruction-detection enables compact, low-latency infrared object detection.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.