首页 >  , Vol. , Issue () : -

摘要

全文摘要次数: 196 全文下载次数: 314
引用本文:

DOI:

10.11834/jrs.20210170

收稿日期:

2020-05-21

修改日期:

2021-01-10

PDF Free   EndNote   BibTeX
基于多尺度深度特征融合网络的遥感图像目标检测
范新南, 严炜, 史朋飞, 张学武
河海大学 物联网工程学院
摘要:

本文针对现有方法对遥感图像目标检测准确率低的问题,在更快速区域卷积神经网络Faster R-CNN(Faster Region Convolutional Neural Networks)算法的基础上对其进行改进,提出一种新的遥感图像目标检测算法。该算法把Faster R-CNN算法中的VGG(Visual Geometry Group)特征提取网络替换为残差网络ResNet(Residual Networks),在此基础上加入特征金字塔网络以充分表达语义信息和位置信息,并使用焦点损失函数替代Faster R-CNN算法中的交叉熵损失函数以解决难易样本对总损失贡献的权重问题,最后对NWPU VHR-10数据集和RSOD数据集采用数据增广方法以解决数据集中图像样本数量少的问题。为验证本文算法的效果,进行了两组对比实验。第一组实验为本文提出的改进模块在NWPU VHR-10数据集和RSOD数据集上的消融实验;第二组实验为本文算法与其他算法在NWPU VHR-10数据集上的对比实验。实验结果表明,本文算法在NWPU VHR-10数据集和RSOD数据集上的多类平均准确率分别达到93.4%和93.0%,比Faster R-CNN算法提高了10.6%和7.8%,同时也高于现有的其他几种算法。

Remote sensing image target detection based on multi-scale deep feature fusion network
Abstract:

Objective: Some existing target detection algorithms are insufficient for feature extraction in remote sensing images and cannot solve the difficult problem of large target scale differences in remote sensing images, especially in the detection of small targets, which leads to low average detection accuracy. In response to these problems, this paper uses the Faster R-CNN (Faster Region Convolutional Neural Networks) algorithm as the basic algorithm, and combines the target characteristics in the remote sensing image to improve the basic algorithm, and finally proposes a new remote sensing image target detection algorithm. Method: First of all, the ResNet (Residual Networks) with more powerful feature extraction capabilities is used to replace the VGG (Visual Geometry Group) network in the original algorithm to solve the shortcomings of the original algorithm"s insufficient feature extraction of remote sensing images. The deep residual network adopts the identity mapping method, which not only ensures that the performance of the network will not degrade as the network deepens, but also extracts deeper features. Secondly, a feature pyramid network is added to the algorithm to fully integrate feature maps of different scales. The feature map obtained in this way has both high-level semantic information and low-level detail information. Therefore, category information and location information can be taken into account. To a certain extent, it can solve the difficult problem of large target scale differences in remote sensing images, and can improve the detection accuracy of small targets greatly. In addition, the focal loss function is used to replace the cross entropy loss function in the original algorithm to solve the problem of the weight of the hard and easy samples to the total loss. Finally, in view of the problem that the used data set contains a small number of images, the data set is expanded using data augmentation. Result: In order to verify the effect of this algorithm, two sets of comparative experiments were carried out. The first set of experiments is the ablation experiment on the NWPU VHR-10 dataset and RSOD-Dataset of the improved modules proposed in this paper; the second set of experiments is the comparison experiment of the algorithm in this paper and other comparison algorithms on the NWPU VHR-10 dataset. The results of the first set of ablation experiments show that the various improved modules proposed in this paper can help improve the accuracy of target detection in remote sensing images, which also makes the target detection accuracy rates of algorithm in this paper can reach 93.4% and 93.0% on the NWPU VHR-10 dataset and RSOD-Dataset respectively. The results of the second set of comparative experiments show that the target detection accuracy of the proposed algorithm is better than the comparison algorithm, which further proves that the proposed algorithm has good performance in the field of remote sensing image target detection. Conclusion: The remote sensing image target detection algorithm proposed in this paper can better solve the difficult problem of large difference in target scale in remote sensing images, and then can improve the target detection accuracy of remote sensing images, especially the detection accuracy of small targets.

本文暂时没有被引用!

欢迎关注学报微信

遥感学报交流群 分享按钮