摘要
为解决复杂战场环境下军事目标识别漏检、误检率高,精度低等问题,提出一种改进YOLOv8的LBI-YOLO军事目标识别算法。该算法引入大核选择性注意力机制以增强主干网络的特征提取能力,使模型能更好地关注重要区域;再采用BiFPN进行多尺度特征融合及用Inner-IoU损失替代传统IoU,加速模型收敛并提高识别精度。实验结果表明,改进后的算法在自建军事目标数据集下识别的mAP值提升了5.3%,FPS提升7.4%。
To address the issues of high missed detection and false detection rates,as well as lower accuracy of military target recognition in complex battlefield environment,an LBI-YOLO military target recognition algorithm with improved YOLOv8 is proposed.The algorithm introduces a large-kernel selective attention mechanism to enhance the feature extraction capability of the backbone network,enabling the model to better focus on important areas;BiFPN is used for multi-scale feature fusion and Inner-IoU loss is use to replace traditional IoU to accelerate model convergence and improve recognition accuracy.Experimental results show that the improved algorithm increase the mAP value of recognition by 5.3%and FPS by 7.4%in the self-built military target dataset.
作者
程国建
沈守婷
白俊卿
CHENG Guojian;SHEN Shouting;BAI Junqing(School of Computer Science,Xi’an Shiyou University,Xi’an 710000,China)
出处
《火力与指挥控制》
北大核心
2025年第8期47-55,共9页
Fire Control & Command Control
基金
陕西省自然科学基金基础研究计划(2023-JC-YB-601)
陕西省计算机学会&翔腾公司基金资助&西安市科技计划高校院所人才服务企业项目(23GXFW0077)。
