一种基于改进的DCL算法与SSD检测网络的遥感影像检测方法

Research on Remote Sensing Image Detection Based on Improved DCL Algorithm

大尺寸遥感图像中小目标和密集目标的检测一直存在着检测准确率低、漏检率高的问题。基于此,以改进的DCL算法处理遥感影像,实现了区块划分,再对各区块的显著性区域进行提取。同时利用深度学习SSD网络,针对密集小目标,进行了目标数量及间距占比研究,并得出定量关系。实验结果显示该方法的检测准确率大大提高,漏检率大大降低,检测时间明显缩短,成功实现了大尺寸遥感影像中的目标自动识别问题。

The detection of small targets and dense targets in large-scale remote sensing images has always had the problems of low accuracy of detection and high rate of missed detection.To overcome these problems,the remote sensing image is processed with the improved DCL algorithm to achieve block partitioning,and then the salient regions of each block are extracted.At the same time,using deep-learning SSD network,the number of targets and the proportion of pitches are studied for small and dense targets,and a quantitative relationship iss obtained.The experimental results show that the detection accuracy of this method is greatly improved,the missed detection rate is greatly reduced,the detection time is significantly shortened,and the automatic target recognition problem in large-size remote sensing images is successfully achieved.