基于YOLOv5s 6.0的飞机遥感图像分类研究  

Research on Aircraft Remote Sensing Image Classification Based on YOLOv5s 6.0

作　　者：顾旭璐 郭中华[1,2] 闫梓旭 陈旺 龚轩 王正 GU Xulu;GUO Zhonghua;YAN Zixu;CHEN Wang;GONG Xuan;WANG Zheng(School of Electronic and Electrical,Ningxia University,Yinchuan 750021,China;Ningxia Key Laboratory of Desert Information Intelligence Perception,Yinchuan 750021,China;Ningxia Institute of Metrology,Quality Inspection and Testing,Yinchuan 750001,China;Guohua Ningxia New Energy Co.,Ltd.,Yinchuan 750000,China)

机构地区：[1]宁夏大学电子与电气工程学院,宁夏银川750021 [2]宁夏沙漠信息智能感知重点实验室,宁夏银川750021 [3]宁夏计量质量检验检测研究院,宁夏银川750001 [4]国华宁夏新能源有限公司,宁夏银川750000

出　　处：《宁夏工程技术》2023年第2期187-192,共6页Ningxia Engineering Technology

基　　金：宁夏自然科学基金项目(2020AAC03026)。

摘　　要：为了实现光学遥感图像中飞机目标的快速分类,提出了以YOLOv5s 6.0为基础的网络模型算法。该算法采用自适应锚框计算得到适用于实验中自建数据集的锚框;使用Mosaic数据增强方法增强了数据集的丰富度;使用SPPF模块提高了计算速度;在Backbone部分引入了注意力机制模块CBAM,在Neck部分引入了加权双向特征金字塔网络BiFPN,以便于将高层次与低层次的特征融合,同时保留了原始特征信息,提高了算法的检测分类能力。最后,根据飞机外型特征点,在自建的3种分类数据集上进行了训练。结果表明:平均检测精度的均值达到60.3%,较改进前的算法模型提高了2.9%;该算法可以实现在一张遥感图像上对飞机目标进行快速定位与分类。In order to achieve rapid positioning and classification of aircraft targets in optical remote sensing,a network model algorithm based on YOLOv5s 6.0 was proposed.It used adaptive anchor frame calculation to obtain the anchor frame suitable for the self-built datasets in experiments,Mosaic data method to enhance the richness of datasets,and SPPF to improve the calculation speed.In addition,it introduced the attention mechanism module CBAM into the Backbone part,and the weighted bi-directional feature pyramid network BiFPN into the Neck section,which combined the high-level and low-level features,retained the information of the original features,and improved the detection and classification ability of the algorithm.The training was conducted on three self-built classification datasets based on the aircraft appearance feature points.The results show that the average detection accuracy reaches 60.3%,which is 2.9%better than the original algorithm model.The algorithm can effectively improve the rapid positioning and classification of aircraft targets on a remote sensing image.

关 键 词：飞机遥感图像分类 YOLOv5s 6.0 BiFPN 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]