基于多尺度融合特征卷积神经网络的遥感图像飞机目标检测

Aircraft detection in remote sensing imagery with multi-scale feature fusion convolutional neural networks

飞机检测在遥感图像解译中具有重要的研究意义。针对现有目标检测算法对于复杂场景区域或飞机密集区域的小尺度飞机目标检测精度较低的问题,本文提出了一种端到端的多尺度特征融合飞机目标检测框架MultDet。该方法基于SSD多尺度检测框架,采用轻量级基础网络提取多尺度特征信息;然后设计反卷积特征融合模块,通过跳跃连接将高层语义特征与低层细节特征进行特征融合,得到结构层次丰富的多尺度融合特征;最后设计了一系列不同纵横比的候选框以适应多尺度飞机目标检测。本文在光学遥感图像数据集UCAS-AOD上进行数据分析试验,结果表明,MultDet512在飞机数据集上取得了94.8%的平均检测精度(average precision,AP),在Titan Xp GPU上达到0.0500s/img的检测速度。本文所提飞机目标检测算法在包含多种复杂场景的遥感图像中,能够实现多尺度飞机目标的高精度稳健检测。

Aircraft detection in remote sensing images (RSIs) is a meaningful task. There are many problems in current detection methods, such as low accuracy in complex background and dense aircraft area, especially for small-scale aircraft. To solve these problems, an end-to-end aircraft detection method named MultDet is proposed in this paper. Based on single shot multibox detector (SSD), a lightweight baseline Network is used to extract multi-scale features for its powerful ability in feature extraction. To obtain the feature maps with enriched representation power, then the multi-scale deconvolution feature fusion block is designed. We add the high-level features with rich semantic information to the low-level features via deconvolution fusion block. In order to locate aircraft of various scales more accurately, a series of aspect ratios of default boxes are set to better match aircraft shapes and combine predictions deduced from feature maps of different layers. The quantitative comparison analysis are carried out on the challenging UCAS-AOD data set. The experimental results demonstrate that the proposed method is accurate and robust for multi-scale aircraft detection, and achieves 94.8% AP(average precision) at the speed of 0.050 0 s/img with the input size 512×512 using a single Nvidia Titan Xp GPU.