Dual-Integrated Low-Latency Single-Lens Infrared Computational Imaging for Object Detection

2026-05-22 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Internet of Things (IoT) & Connected Devices · Depth: Expert, extended

Summary

The Physics-aware Dual-Integrated Network (PDI-Net) introduces a novel approach for low-latency single-lens infrared computational imaging, addressing the trade-off between accuracy and inference speed in compact systems. This network integrates physical priors from the optical path with deep learning, combining infrared image reconstruction and object detection. PDI-Net utilizes a U-Net for reconstruction, with a semi-U-Net acting as a feature-sharing layer connected to a YOLO backbone. A Physics-aware Large-Small Bridge (PALS-Bridge) aligns fidelity-oriented features from the semi-U-Net with detection-oriented semantic representations in YOLO, adaptively modulating multiscale convolutional branches using Point Spread Function (PSF) priors. Deployed on a single-lens infrared camera with an RK3588 edge AI chip, PDI-Net reduces system weight by approximately 50% (from 700g to 372g). It also cuts inference time by 84.06% to 5.52 ms and improves mAP@0.5:0.95 by 5.07% compared to a pruned Rec+Det strategy in low-SNR conditions, enabling real-time object detection on UAVs.

Key takeaway

For Machine Learning Engineers deploying real-time object detection on resource-constrained edge devices with infrared cameras, PDI-Net offers a compelling solution. Its integration of optical physics priors and joint reconstruction-detection significantly reduces system weight by 50% and inference latency by 84.06% while boosting detection accuracy (mAP@0.5:0.95 by 5.07%). You should explore physics-aware, dual-integrated architectures to optimize performance on compact, low-power platforms.

Key insights

Integrating optical physics priors with deep learning for joint reconstruction-detection enables compact, low-latency infrared object detection.

Principles

• Physical priors guide deep learning for accuracy-latency balance.
• Joint optimization enhances task-relevant feature discriminability.
• Bypass full reconstruction for significant latency reduction.

Method

PDI-Net uses a U-Net encoder for shared features, PALS-Bridge with PSF priors to align them, and a YOLO backbone, bypassing full reconstruction during inference.

In practice

• Reduce system weight by 50% with single-lens designs.
• Achieve 84% faster inference with PDI-Net on edge devices.
• Modulate feature extraction using PSF priors for blur adaptation.

Topics

• Infrared Imaging
• Computational Imaging
• Object Detection
• Deep Learning
• YOLO Architecture
• Physics-aware AI
• Edge Deployment

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

Related on AIssential

• Learning with Semantic Priors: Stabilizing Point-Supervised Infrared Small Target Detection via Hierarchical Knowledge Distillation — Hierarchical knowledge distillation with VFM semantics stabilizes point-supervised infrared small target detection.
• YOLO — You Only Look Once — A decade of real-time vision — YOLO transformed object detection into a single regression problem for real-time performance.
• 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.
• FATE: Pillar Encoding and Frequency-Aware Training for Event-Based Object Detection — FATE uses Pillar Encoding and Frequency-Aware Training to enable robust, high-frequency object detection from sparse event camera data.
• Fusing Transferred Priors and Physics-based Decomposition for Underwater Image Enhancement — A novel UIE method fuses transferred priors and physics-based decomposition to enhance underwater images without paired labels.
• DualGate-Net: A Prior-Gated Dual-Encoder Framework for Histopathology Cell Detection — Adaptive, gated fusion of local and global features with contextual priors improves histopathology cell detection.

Open in AIssential →

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