10K is Enough: An Ultra-Lightweight Binarized Network for Infrared Small-Target Detection

2025-03-04

AI Summary

To address the accuracy-efficiency trade-off in deploying infrared small target detection (IRSTD) on resource-constrained edge devices, this paper proposes BiisNet—a super-lightweight binarized network with only ~10K parameters. To mitigate accuracy degradation inherent to binarization, we introduce three key innovations: (1) Dot-Binary Convolution, which jointly leverages binary computation and full-precision feature streams; (2) a Dynamic Softsign function enabling smooth, adaptive gradient optimization and improved weight distribution; and (3) a feature-level precision preservation mechanism. Evaluated on mainstream IRSTD benchmarks, BiisNet substantially outperforms all existing binarized methods, achieving accuracy comparable to state-of-the-art full-precision models while accelerating inference by over 5×. This work establishes a new paradigm for high-accuracy, ultra-low-power IRSTD deployment on edge devices.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Computer Vision: Biometrics, Face, Gesture & Pose
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

The widespread deployment of InfRared Small-Target Detection(IRSTD) algorithms on edge devices necessitates the exploration of model compression techniques. Binary neural networks (BNNs) are distinguished by their exceptional efficiency in model compression. However, the small size of infrared targets introduces stringent precision requirements for the IRSTD task, while the inherent precision loss during binarization presents a significant challenge. To address this, we propose the Binarized Infrared Small-Target Detection Network (BiisNet), which preserves the core operations of binarized convolutions while integrating full-precision features into the network's information flow. Specifically, we propose the Dot-Binary Convolution, which retains fine-grained semantic information in feature maps while still leveraging the binarized convolution operations. In addition, we introduce a smooth and adaptive Dynamic Softsign function, which provides more comprehensive and progressively finer gradient during back-propagation, enhancing model stability and promoting an optimal weight distribution.Experimental results demonstrate that BiisNet not only significantly outperforms other binary architectures but also demonstrates strong competitiveness among state-of-the-art full-precision models.

Problem

• Explores model compression for edge-deployed IRSTD algorithms.
• Addresses precision loss in binarized networks for small infrared targets.
• Proposes BiisNet with Dot-Binary Convolution and Dynamic Softsign function.

Innovation

• Binarized Infrared Small-Target Detection Network (BiisNet)
• Dot-Binary Convolution retains fine-grained semantic information
• Dynamic Softsign function enhances model stability