基于改进YOLOv3的机场盲区遥感图像目标检测方法

针对机场存在的雷达监视盲区问题,提出一种改进YOLOv3目标检测算法。首先，基于原YOLOv3主干特征提取网络加入SPP池化模块，以深度可分离卷积替代普通卷积。然后，针对小尺度目标数据集，增加第4层金字塔加强特征提取网络，在kmeans++聚类算法的基础上提出一种线性放缩进行锚框筛选。最后，在RSOD-Dataset数据集上MAP为88.82%，FPS为37.57。仿真结果表明该方法可以满足机场实时目标检测任务的需求。

Remote Sensing Images Target Detection Method Based on Improved YOLOv3 for Blind Areas of Airports

An improved YOLOv3 target detection algorithm is proposed to address the problem of radar surveillance blind areas in airports. Firstly, based on the original YOLOv3 backbone feature extraction network, an SPP pooling module is added to replace the normal convolution with a depth-separable convolution.Then, for the small-scale target dataset, a fourth layer pyramid is added to strengthen the feature extraction network, and a linear deflation is proposed for anchor frame screening based on the kmeans++ clustering algorithm. Finally, the MAP value on the RSOD-Dataset dataset is 88.82% and the FPS is 37.57. The simulation results show the effectiveness of the algorithm can meet the requirements of real-time target detection tasks in airports.