基于Sparse R-CNN的遥感目标检测研究
Research on remote sensing object detection based on sparse R-CNN
刘冰 1段睿1
作者信息
- 1. 长春工业大学计算机科学与工程学院,吉林长春 130102
- 折叠
摘要
遥感图像目标检测任务在天气预报、环境监测及军事应用等领域均有应用,但其小目标众多、类间相似度大、尺度多样等问题导致提取特征困难.基于深度学习的方法在目标检测领域已经流行起来,Sparse R-CNN是一种结构简单且效果较好的模型,但将其直接应用到遥感图像上结果较差,针对遥感图像特点引入了自监督学习框架跟选择性查询收集提高了遥感图像目标检测的效果,在mAP指标上提高3.8个百分点.
Abstract
Remote sensing image object detection tasks have applications in fields such as weather forecasting,environmental monitoring,and military applications.However,the numerous small targets,high similarity between classes,and diverse scales make it difficult to extract features.The method based on deep learning has become the mainstream in the field of object detection.Sparse R-CNN is a model with simple structure and good effect,but its direct application to remote sensing images has poor results.This paper introduces a Self-Supervised learning framework and selective query recollection according to the characteristics of remote sensing images to improve the effect of target detection in remote sensing images,improved by approximately 3.8 percentage points on the mAP metrics.
关键词
遥感图像/目标检测/自监督/基于查询的目标检测方法Key words
remote sensing image/object detection/self-supervised learning/query-based object detection引用本文复制引用
基金项目
吉林省教育厅基金资助项目(JJKH20230765KJ)
出版年
2024
NETL
NSTL
万方数据