基于多感知融合的遥感影像检测算法
Remote Sensing Image Detection Algorithm Based on Multi-sensory Fusion
何中良 1赵良军 1宁峰 2席裕斌 1梁刚1
作者信息
- 1. 四川轻化工大学计算机科学与工程学院,四川宜宾 643002
- 2. 四川轻化工大学自动化与信息工程学院,四川宜宾 643002
- 折叠
摘要
针对遥感影像复杂背景和小目标检测困难的问题,提出了一种基于多感知融合的检测算法YOLO-GT.为了提升特征图中小目标的特征信息,设计了包含3种感知机制的检测头Adaptive Scale-Aware Dynamic Head(ASADH);引入轻量级上采样算子Content-Aware ReAssembly of Features(CARAFE),解决语义信息丢失问题,提升特征金字塔网络性能;为进一步优化模型的训练速度和定位精度,采用了 Wise-IoU作为损失函数.实验结果在DIOR数据集上显示,模型精度达90.4％,比原算法提高2.1％.这些改进有效提高了复杂背景下遥感影像小目标的检测性能.
Abstract
To deal with the problem of complex background and small target detection in remote sensing images,a multi-sensory fusion-based detection algorithm YOLO-GT is proposed.In order to improve the feature information of small targets in the feature map,a detection head Adaptive Scale-Aware Dynamic Head(ASADH)which contains three kinds of sensing mechanisms is designed;at the same time,a lightweight up-sampling operator Content-Aware ReAssembly of Features(CARAFE)is introduced to solve the problem of semantic information loss and improve the performance of the feature pyramid network;then Wise-IoU is adopted as the loss function in order to further optimize the training speed and localization accuracy of the model.The experimental results on the DIOR dataset show that the model accuracy reaches 90.4％,which is 2.1％higher than the original algorithm.These improvements effectively enhance the performance of small targets detection in remote sensing images under complex backgrounds.
关键词
遥感影像/目标检测/多感知融合/深度学习Key words
remote sensing image/target detection/multi-sensory fusion/deep learning引用本文复制引用
基金项目
四川省科技计划项目(2023YFS0371)
四川省智慧旅游研究基地项目(ZHZJ22-03)
出版年
2024
国家科技期刊平台
NSTL
万方数据