期刊文献+

轻量级卷积网络遥感影像飞机目标检测方法研究

Research on aircraft target detection method in lightweight convolutional network remote sensing images
下载PDF
导出
摘要 针对常规级遥感影像飞机检测模型难以在低功耗硬件环境部署的问题,提出一种轻量级检测模型,以分组卷积特征提取层,利用轻型通道注意力模块提高对正样本的学习能力,通过三层多尺度特征金字塔增强高层特征图中的语义信息。利用多个来源的遥感影像构建数据集,并对训练集进行质量与样本增强。使用大型计算机训练模型,然后将模型迁移至低功耗硬件中完成测试。结果表明:本文提出模型在测试环境下达到每张0.087 s的检测速度,训练后模型仅占43.41 MB,同时具备较好的泛化能力,在精度、速度和模型体量几方面均优于对比模型,说明本文提出模型能够部署于低功耗硬件终端并对飞机目标实施快速精准地检测。 Aiming at the problem that the conventional remote sensing image aircraft detection model is difficult to deploy in a lowpower hardware environment,a lightweight detection model is proposed,which uses grouped convolution feature extraction layers and uses a lightweight channel attention module to improve the learning ability of positive samples.The complexity of semantic information in the output feature map is improved through a three-layer multi-scale feature pyramid.Use remote sensing images from multiple sources to construct a data set,and perform quality and sample enhancement on the training set.Use large computers to train models and then move the models to low-power hardware for testing.The results show that the model proposed in this article achieves a detection speed of 0.087 seconds/photo in the test environment.After being trained,the model only occupies 43.41MB of memory.It also has good generalization ability and is better than the previous model in terms of accuracy,speed and model size.Comparing the models shows that the model proposed in this article can be deployed on low-power hardware terminals and perform fast and accurate detection of aircraft targets.
作者 都凯 DU Kai(Yuanqu County Surveying and Mapping Geographic Information Center,Yuanqu 043700,China)
出处 《经纬天地》 2023年第6期18-23,共6页 Survey World
关键词 遥感影像 飞机检测 轻量化模型 轻型注意力机制 remote sensing images aircraft inspection lightweight model light attention mechanism
  • 相关文献

参考文献5

二级参考文献18

共引文献61

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部