Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (17): 232-241.DOI: 10.3778/j.issn.1002-8331.2205-0058

• Graphics and Image Processing • Previous Articles     Next Articles

Improved Feature Fusion Network for Small Object Detection in Remote Sensing Images

LI Chao, WANG Kai, DING Caichang, ZHANG Jinyue, LI Jiabao   

  1. 1.School of Computer, Hubei University of Technology, Wuhan 430000, China
    2.School of Computer and Information Science, Hubei Engineering University, Xiaogan,Hubei 432000, China
  • Online:2023-09-01 Published:2023-09-01

改进特征融合网络的遥感图像小目标检测

李超,王凯,丁才昌,张金玥,李贾宝   

  1. 1.湖北工业大学 计算机学院,武汉 430000
    2.湖北工程学院 计算机与信息科学学院,湖北 孝感 432000

Abstract: Small scale and dense arrangement of objects in remote sensing images present significant challenges for feature extraction and object detection. To solve the challenges above, a network based on feature fusion and attention mechanism is proposed. Firstly, to enhance detailed features which are useful for small scale object, a scale attention module is proposed to filter invalid semantic features and generate attention masks with different scales after the backbone network. Moreover, an edge refinement module is introduced to reduce false detection by suppressing feature misalignment in the dense area. Experimental results show that, compared with the baseline model Faster R-CNN, the method improves the detection accuracy AP50 and APS by 10 and 11.1 percentage points, respectively.

Key words: remote sensing image, small object detection, attention mechanism, feature fusion

摘要: 遥感图像中目标尺寸小且密集排布的特点给特征提取和目标检测带来了困难。针对上述问题,提出一种基于注意力机制的特征融合网络ABFN(attention based feature fusion network)。通过尺度注意力模块在骨干网络提取的特征图上生成不同尺度的注意力Mask过滤无效的语义信息,以此增强小目标的细节信息;引入边缘细化模块抑制目标密集区域的特征错位,以此降低目标的误检率。实验结果表明,相比于基准模型Faster R-CNN,该方法在遥感数据集xView上的检测精度AP50和APS分别提高了10和11.1个百分点。

关键词: 遥感图像, 小目标检测, 注意力机制, 特征融合