基于多尺度特征稠密连接的遥感图像目标检测方法

Research on Remote Sensing Image Object Detection Method Based on Densely Connected Multi-scale Features

遥感图像目标检测是指在遥感图像中快速识别并定位感兴趣的各类目标,是遥感图像解译的基础性工作。传统的目标检测方法主要基于人工设计特征或浅层学习特征,由于特征表征能力有限,检测效果并不佳。深度学习技术的蓬勃发展为遥感图像目标检测提供了一个非常有效的工具,但是由于遥感图像的特殊性,目前主流的基于深度学习的通用目标检测方法无法直接用于遥感图像目标检测中。针对这些问题,首先分析了遥感图像数据的固有特点和新的分析应用需求,然后详细介绍了基于深度学习的目标检测算法YOLO,并在此基础上对算法进行了针对性改进,提出一种基于多尺度特征稠密连接的遥感图像目标检测方法,实验结果表明,提出的方法在相关数据集上检测的平均精度相比于YOLOv3提高了1. 91个百分点,具有相对较好的检测效果。

Remote sensing image object detection refers to the rapid identification and location of various types of objects in remote sensing images,which is the basic work of remote sensing image interpretation.The traditional object detection method is mainly based on artificial design features or shallow learning features. Due to limited feature representation,the result is not good. The fast development of deep learning technology provides a very effective tool for remote sensing image object detection. However,due to the particularity of remote sensing images,the current generic object detection method based on deep learning cannot be directly used in remote sensing image object detection. This paper first analyzes the inherent characteristics of remote sensing image data and the new analytical application requirements,then introduces an object detection algorithm based on deep learning YOLO in detail,and then,an object detection method based on densely connected multi-scale features is proposed. The experimental show that the method proposed in this paper improves the Average-Precision of the detection result on the relevant dataset by about 3-4 percentage compared with YOLOv3.