基于深度学习的光学遥感图像目标检测综述

Survey of object detection in optical remotesensing images based on deep learning

目标检测是光学遥感图像解译的核心环节,广泛应用于情报侦察、土地利用、城市规划等领域。近年来,深度学习的发展成熟使光学遥感目标检测的精确度和效率得到大幅提升。本文首先综述了基于深度学习的通用目标检测算法;然后介绍了当前常用的光学遥感图像目标检测数据集并依据数据特点分析了数据集的发展趋势;接着根据遥感图像检测难点,从任意方向、多尺度、小目标、密集分布、复杂背景5个方面详细梳理了算法的优化方案;最后展望了光学遥感图像目标检测研究的发展方向。

Object detection is the core part of optical remote sensing image interpretation,which has wide applications in intelligence reconnaissance,land resource utilization,urban planning and other fields.In recent years,the maturation of deep learning has led to significant improvements in the accuracy and efficiency of optical remote sensing object detection.In this paper,the general target detection algorithms based on deep learning is reviewed first.Secondly,the current commonly used optical remote sensing image object detection data set is introduced and the development trend of the data set based on the data characteristics is analyzed.Then,according to the difficulties of remote sensing image detection,the optimization scheme of the algorithm is sorted out in detail from five aspects:arbitrary direction,multi scale,small target,dense distribution,and complex background.Finally,the development direction of optical remote sensing image object detection research is prospected.