光学学报, 2018, 38 (7): 0728001, 网络出版: 2018-09-05
改进区域卷积神经网络的机场检测方法 下载: 959次
Airport Detection Method with Improved Region-Based Convolutional Neural Network
遥感 遥感图像 机场检测 卷积神经网络 交叉优化 remote sensing remote sensing image airport detection convolutional neural network alternating optimization
摘要
提出了一种结合级联的区域建议网络和检测网络的遥感图像机场检测方法。通过改进区域建议网络,以获得高质量的机场建议框;通过改进检测网络的损失函数,以提高机场检测的准确性;使用交叉优化策略,实现了两个网络的卷积层共享,机场检测时间大幅缩减。结果表明,所提方法在复杂背景下能准确地检测出不同类型的机场,检测率高,虚警率低,平均运行时间短。
Abstract
An airport detection method based on remote sensing images which combines the cascaded regional proposal network with the detection network is proposed. The regional proposal network is improved to get the airport proposal boxes with a high quality, and the loss function of the detection network is improved to increase the accuracy of the airport detection. In addition, the alternating optimization strategy is adopted to share the convolution layers between the two networks, and thus the airport detection time is greatly shortened. The results show that this proposed method can be used to accurately detect different types of airports under complex backgrounds with a high detection rate, a low false-alarm rate and short average processing time.
朱明明, 许悦雷, 马时平, 唐红, 辛鹏, 马红强. 改进区域卷积神经网络的机场检测方法[J]. 光学学报, 2018, 38(7): 0728001. Mingming Zhu, Yuelei Xu, Shiping Ma, Hong Tang, Peng Xin, Hongqiang Ma. Airport Detection Method with Improved Region-Based Convolutional Neural Network[J]. Acta Optica Sinica, 2018, 38(7): 0728001.