Infrared dim-small target detection algorithm based on multiple perception fields
基于多感受野的红外弱小目标检测算法

摘要

针对红外图像中弱小目标像素占比低、特征稀疏、背景干扰严重等问题，提出一种基于多感受野特征融合的红外弱小目标检测算法。该算法以YOLOv5s为基础检测框架，通过设计多分支感受野扩展模块，在颈部网络中构建多尺度上下文信息融合机制，增强了网络对弱小目标的特征表达能力；同时引入自适应特征加权策略，有效抑制复杂背景噪声干扰。在公开红外弱小目标数据集上的实验结果表明，改进后算法相较于基线YOLOv5s，在检测精度、召回率及定位准确性等指标上均有显著提升，验证了多感受野融合策略对红外弱小目标检测任务的有效性。

Abstract

To address the challenges of extremely low pixel occupancy, sparse features, and severe background interference in infrared dim and small target detection, this paper proposes an infrared dim and small target detection algorithm based on multi-receptive field feature fusion. The algorithm adopts YOLOv5 as the baseline detection framework and constructs a multi-scale contextual information fusion mechanism within neck networks by designing a multi-branch receptive field expansion module, thereby enhancing the network's feature representation capability for dim and small targets. Meanwhile, an adaptive feature weighting strategy is introduced to effectively suppress interference from complex background clutter. Experimental results on public infrared dim and small target datasets demonstrate that, compared with the baseline YOLOv5, the improved algorithm achieves significant improvements in detection precision, recall, and localization accuracy, validating the effectiveness of the multi-receptive field fusion strategy for infrared dim and small target detection tasks.