基于改进SSD算法的遥感图像目标检测

在遥感图像目标检测领域，多数目标检测算法针对小目标检测时效果不佳，为此，提出一种多尺度特征融合的遥感图像目标检测算法。利用SSD算法的基础网络进行特征提取，形成特征图金字塔。设计特征图融合模块，融合浅层特征图的位置信息和深层特征图的语义信息，从而保留丰富的上下文信息。设计冗余信息去除模块，通过卷积操作进一步提取特征图中的特征，并对特征信息进行筛选，以减少特征图融合时带来的混叠效应。在遥感图像数据集NWPU VHR-10上的实验结果表明，该算法的平均检测精度高达93.9%，其针对遥感图像小目标的检测性能优于Faster R-CNN和SSD等算法。

Object Detection in Remote Sensing Images Based on Improved SSD Algorithm

In the field of object detection in remote sensing images, most of the existing object detection algorithms perform poorly for small objects.This paper proposes an algorithm that fuses multi-scale features for object detection in remote sensing images.The features are first extracted by using the basic network of the SSD algorithm to form a feature map pyramid.Then the feature map fusion module is designed to fuse the position information of the shallow feature map and the semantic information of the deep feature map, retaining rich context information.Finally, a module to remove redundant information is designed, and the features in the feature map are further extracted through the convolution operation.The feature information is also screened to reduce the aliasing effect brought by the fusion of the feature maps.The experimental results on NWPU VHR-10, a dataset of remote sensing images, show that the proposed algorithm achieves an average detection accuracy of 93.9%, demonstrating that it outperforms Faster R-CNN, SSD and other algorithms in detection of small objects in remote sensing images.