Abstract: To deal with large amount of remote sensing image data and the shortcoming of existing airport detection methods, a new algorithm which eliminates repetitive patterns, fusing linear saliency and regional saliency is proposed. Firstly, an improved saliency model based on Hypercomplex Fourier Transform（HFT） is used to remove repetitive patterns in remote sensing images, so as to reduce the amount of subsequent data processing. Secondly, runways have parallel long straight line characteristics and there are differences between the airport and the surrounding environment. The linear saliency map and the regional saliency map are calculated according to these characteristics. Then two saliency maps are fused. Finally, the candidate regions are determined according to the runway width. The deep Convolution Neural Network（CNN） and Support Vector Machine（SVM） are used to extract feature and determine whether the regions contain airports or not. Experimental results show that this algorithm is more accurate for the airport location, with higher recognition rate, lower false alarm rate and faster speed.

Key words: airport detection, repetitive patterns, linear parallelism, local region contrast, deep convolution neural network

摘要： 针对遥感图像数据量大和现有机场检测方法存在的不足，提出了一种去除重复模式，融合直线显著性和区域显著性的机场检测方法。首先利用改进的基于超复数傅里叶变换的显著性模型，去除遥感图像中的重复模式，减少后续数据处理量；然后根据跑道平行长直线特性和机场与周围环境的差异性，计算基于直线和基于区域的显著图，并进行融合；最后结合跑道宽度确定候选区，通过深度卷积神经网络和支持向量机进行特征提取和识别。实验证明所提算法对机场定位更加准确，具有识别率高、虚警率低、速度快的特点。

关键词: 机场检测, 重复模式, 直线平行性, 局部对比度, 深度卷积网络