基于DFECANet的遥感图像飞机目标检测方法

针对现有的遥感图像目标检测方法中对小尺寸飞机目标的检测精度不高、特征信息传递不准确、信息交互不充分等问题，提出了一种基于可辨别特征提取和上下文感知的遥感图像飞机目标检测方法。设计了以可辨别特征提取模块为主体的主干网络，用以加强对多尺度飞机目标的特征提取；引入自适应特征增强模块，选择性关注小目标、优化特征信息的传递与信息交互；并设计了特征融合上采样模块对特征图进行上采样操作，用以提升高层语义信息的准确性。在DOTAv1数据集上的检测精度达到了95.2%,相较于YOLOv5s、SCRDet、ASSD等主流算法，飞机目标的检测精度提高了3.7%～18%。此外，该方法的检测速度以及模型参数量分别为147 fps和13.4 M,相较于当前主流算法具备较强的竞争力，满足在遥感背景下对飞机目标的实时检测需求。

Aircraft target detection in remote sensing images based on DFECANet

Aiming at the existing remote sensing image target detection methods with low detection accuracy for small-size aircraft targets, inaccurate feature information transfer and insufficient information interaction, a remote sensing image aircraft target detection method based on discriminative feature extraction and context-awareness is proposed. A backbone network with a discriminative feature extraction module is designed to enhance feature extraction for multi-scale aircraft targets; an adaptive feature enhancement module is introduced to selectively focus on small targets and optimize the transfer of feature information and information interaction; and a feature fusion up-sampling module is designed to perform up-sampling operations on the feature maps to improve the accuracy of high-level semantic information. The detection accuracy on the DOTAv1 dataset reaches 95.2%, which is 3.7% to 18% higher than that of mainstream algorithms such as YOLOv5s, SCRDet, ASSD. In addition, the detection speed and the number of model parameters of the proposed method are 147 frames per second and 13.4 M, respectively. Compared with the current mainstream algorithms, the proposed method has strong competitiveness and meets the real-time detection requirements of aircraft targets in the background of remote sensing.