基于深度学习的SAR图像道路识别新方法

A new deep learning method for roads recognition from SAR images

针对传统合成孔径雷达(SAR)图像道路识别步骤繁杂的问题,提出了一种新的基于深度学习的SAR图像道路识别方法。首先,在原有全卷积神经网络(FCN)的基础上通过改进激活函数构造一种新的卷积神经网络M-FCN,有效缓解了道路信息丢失问题;然后,将该卷积神经网络和自主构建的道路标签集应用于模拟SAR和真实SAR图像道路识别实验中,提高了鲁棒性。实验结果表明:与支持向量机(SVM)、传统全卷积神经网络和其他算法比较,该算法可以用来识别道路特征,并具有较高的精度和可靠性。

Deep learning is an effective technical method to enhance the recognition accuracy of remote sensing image target. To solve the problem of the complicated steps in road recognition from Synthetic Aperture Radar(SAR)images,this paper proposes a SAR image road recognition method based on deep learning. First,based on the traditional Full Convolution Neural Network(FCNN),a New Convolution Neural Network(NCNN)is constructed by revising the activation function to effectively alleviate the loss of road information. Then,the NCNN is applied to road recognition experiments of simulated SAR images and real SAR images with self-constructed road label sets to improve the robustness of the proposed method. The experimental results show that the NCNN can be used to identify the overall characteristics of the road with higher accuracy and better reliability in comparison with the Support Vector Machine(SVM),traditional FCNN and other algorithms.