首页 >  , Vol. , Issue () : -

摘要

全文摘要次数: 124 全文下载次数: 139
引用本文:

DOI:

10.11834/jrs.20210507

收稿日期:

2020-11-09

修改日期:

2021-09-13

PDF Free   EndNote   BibTeX
基于特征注意力金字塔的单阶段遥感图像目标检测方法
摘要:

遥感图像场景复杂、目标大小不一、分布不均衡等特点增加了目标检测的难度,而适于检测不同尺度目标的特征金字塔融合不同深度的特征图时,没有考虑特征图各自的重要性,更没有强调目标区域的特征,为此本文提出基于特征注意力金字塔的单阶段遥感图像目标检测方法FAPNet(Feature Attention Pyramid Network)。首先,使用通道拼接方式融合不同深度的特征图,给用于检测的特征图提供不同大小感受野的特征,并基于通道注意力对融合的特征图在通道维度重标定,根据特征图所负责检测目标的尺度自适应地调整不同大小感受野特征的权重,强化感受野大小与待检测目标尺度匹配度较高的特征,弱化匹配度较低的特征。其次,使用叠加的扩张空间金字塔池化结构,结合弱监督分割网络建模位置注意力,强化目标区域特征,弱化背景区域特征,进一步提升目标检测方法的性能。实验结果表明,相较于RetinaNet,针对汽车目标,所提方法在UCAS-AOD数据集和RSOD数据集上检测精度AP分别提升了3.41%和2.26%,针对多类目标所提方法在各目标上取得了较优的AP结果,且mAP结果优于其他比较方法。

Feature attention pyramid network for single-stage remote sensing object detection
Abstract:

Abstract: Objective The characteristics of remote sensing images, such as complex scene, different sizes of targets and unbalanced distribution, increase the difficulty of target detection. However, feature pyramids that suitable to detect the targets of different scales do not take into account the importance of different feature maps when fusing the feature maps, let alone to emphasize the features of target areas. For this purpose, this paper proposes a feature attention pyramid based remote sensing image object detection method (namely Feature Attention Pyramid Network, FAPNet). Method Firstly, the feature maps of different depths are fused by channel concatenation, and the features of different sizes receptive fields are provided for the feature maps used for detection. The channel attention mechanism is used to recalibrate the fused feature maps in the channel dimension. The feature maps from different depths are adaptively adjusted according to the scale of the object to be detected, so as to strengthen the feature that matches highly between the size of the receptive field and the object to be detected, and weaken the feature with a low degree of matching. Secondly, the weakly supervised attention module uses the superimposed atrous spatial pyramid pooling structure and convolutional segmentation module to model spatial attention weights, so as to adjust the feature distribution of the feature map that used for prediction, strengthen the object area feature, weaken the background area feature, this further improves the performance of object detection methods. Result The experimental results show that compared with RetinaNet, the proposed method improves the accuracy (AP) for car targets of 3.41% and 2.26% respectively on UCAS-AOD dataset and RSOD dataset, achieves better AP results on each target for multi-class targets, and is superior to other comparative object detection methods on mAP indicator for multi-targets. Conclusion A feature attention pyramid based remote sensing image object detection method is proposed in this paper. Its contribution lies in the designed feature attention pyramid module and weakly supervised attention module. With the new modules, the proposed method can extract target features more accurately in complex scenes with targets of different sizes by channel attention and spatial attention, thus improves the performance of detection. Experimental results show that the proposed method is superior to the RetinaNet and FAN methods, it is more suitable for remote sensing image object detection tasks with complex scenes and multi-scale 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