复杂环境大场景SAR图像飞机目标快速检测

Fast detection of aircrafts in complex large-scene SAR images

随着人工智能与合成孔径雷达(synthetic aperture radar,SAR)技术的发展,基于卷积神经网络(convolutional neural network,CNN)的SAR图像自动目标识别技术取得了一定的突破.然而,由于飞机自身结构以及SAR成像机制的复杂性,在复杂环境大场景SAR图像中对飞机目标进行快速准确的检测依然存在挑战.为提升算法的检测能力,本文对现有检测算法的处理流程进行了分析与总结,并提出了一种复杂环境大场景SAR图像飞机目标快速检测算法.算法优化了整体检测流程,设计了基于灰度特征的机场区域精细化提取和基于CNN的飞机目标粗检测两大子模块,并采用了YOLOv3网络对机场区域以及飞机目标分别进行初步的提取与检测.实验结果表明,本文算法对复杂环境大场景SAR图像中的飞机目标具有高效的检测能力.

With the development of artificial intelligence and synthetic aperture radar(SAR),some breakthroughs have been made in convolutional neural network(CNN)based SAR automatic target recognition(ATR).However,due to the complexity of aircraft’s structures and SAR imaging mechanisms,detecting aircrafts fast and accurately in complex large-scene SAR images is still challenging.To improve detecting performance of algorithms,this paper summarizes the processing flows of current detection algorithms,and proposes a fast detection algorithm for aircrafts in complex large-scene SAR images.The method optimizes the whole processing scheme,designs grayscale based airport refining extraction and CNN based aircraft detection modules as well as using YOLOv3 to extract airport areas and detect aircrafts.Experimental results illustrate that the proposed method could detect aircrafts efficiently in complex large-scene SAR images.