测控技术2025，Vol.44Issue(4)：16-24,9.DOI:10.19708/j.ckjs.2025.04.301
多尺度特征融合的轻量化红外船舶检测算法
Lightweight Infrared Ship Detection Algorithm Based on Multi-Scale Feature Fusion
摘要
Abstract
Identification of maritime ship targets has become a hot research topic in the field of pattern recogni-tion,and detecting ship targets from infrared images holds great potential for a wide range of applications.To address the issues of large model size,parameter redundancy,and low detection efficiency when using the YOLOv7 algorithm for infrared ship detection,a lightweight detection algorithm based on multi-scale feature fu-sion is proposed.Firstly,a reparameterized MobileOne network is introduced to replace the original backbone feature extraction network,which can reduce the model's parameters number.Secondly,the BCBS module is re-constructed to replace the CBS module in the neck network,which can obtain critical ship target information with fewer parameters and faster speed,thus enhancing the model's ability to extract ship feature information.Then,an improved BiFPN-T feature pyramid network model is designed to fuse ship features of different scales and enhance the model's multi-scale fusion capability.Finally,an improved N-CIoU loss function is introduced in the detection head to enhance the detection capabilities in small target scenarios.Experimental results show that the mean average precision(mAP)of the improved algorithm model is increased by 1.6％,the number of parameters is reduced by about 27％,and the detection speed is increased by 30 f/s,which can meet the re-quirements of infrared ship detection for detection accuracy and real-time performance,and demonstrating good detection performance.关键词
红外船舶检测/轻量化/注意力机制/多尺度/损失函数Key words
infrared ship detection/lightweight/attention mechanism/multi-scale/loss function分类
计算机与自动化引用本文复制引用
黄孙港,饶兴昌,李海龙..多尺度特征融合的轻量化红外船舶检测算法[J].测控技术,2025,44(4):16-24,9.基金项目
江苏省教育厅项目(22KJB140016) （22KJB140016）
