基于EC-CSP和双路特征融合的红外无人机目标检测  

Infrared UAV Target Detection Based on EC-CSP and Dual Path Feature Fusion

作　　者：李祥 李昊瞳 周敏敏 LI Xiang;LI Hao-tong;ZHOU Min-min(School of Information and Electrical Engineering,Hebei University of Engineering,Handan 056038,China)

机构地区：[1]河北工程大学信息与电气工程学院,河北邯郸056038

出　　处：《电脑与信息技术》2024年第5期39-43,共5页Computer and Information Technology

基　　金：河北省自然科学基金面上项目(项目编号:F2021402009、A2020402013)。

摘　　要：针对红外无人机目标检测的实时性与准确性需求不断提升的问题,提出基于EC-CSP和双路特征融合的红外无人机目标检测算法。首先,提出结合ECA-CBAM注意力机制的CSP模块(Efficient Channel Attention-Convolutional Block Attention Module-CSP,EC-CSP),使网络能关注更重要的区域。其次,提出包含5×5、9×9像素大小最大池化的STM(Small Target Maxpool)模块,抑制背景特征对小尺度目标特征的干扰;最后,提出融合1×1与3×3基本卷积操作的全局特征提取(Global Feature Extracion,GFE)模块,并与STM模块组成双路特征融合(Dual Path Feature Fuasion,DPFF)模块,提高全局特征和局部特征的融合能力。实验结果表明,新算法取得了良好的实验效果。Aiming at the problem that the real-time and accuracy requirements of infrared UAV target detection are constantly improving,an infrared UAV target detection algorithm based on EC-CSP and dual feature fusion was proposed.Firstly,a CSP module(Efficient Channel Attention-Convolutional Block Attention Module-CSP,EC-CSP)combined with the ECA-CBAM attention mechanism is proposed to enable the network to focus on more important areas.Secondly,a Small target maxpool(STM)module with 5×5 and 9×9 pixel size maximum pooling is proposed to suppress the interference of background features on small-scale target features.Finally,a Global Feature Extracion(GFE)module that integrates 1×1 and 3×3 basic convolution operations is proposed and combined with the STM module to form a Dual Path Feature Fusion(DPFF)module to improve the fusion ability of global features and local features.The experimental results show that the new algorithm has achieved good experimental results.

关 键 词：YOLOv5s 红外无人机 ECA-CBAM EC-CSP 双路特征融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]