Road segmentation algorithm for SAR images based on U-Net and wavelet transform
Traditional SAR image road segmentation is greatly affected by speckle noise,and has diffi-cult in utilizing high-frequency information in the image,and low segmentation accuracy.In response to the above problems,this paper proposes a SAR image road segmentation algorithm based on the wavelet transform attention mechanism and U-Net.A frequency domain attention mechanism based on wavelet transform is designed;a hybrid pooling mechanism is introduced to enhance the elongated features of roads in SAR images;stripe and pyramid hybrid pooling and frequency domain attention are added to U-Net.On this basis,a convolutional neural network for road segmentation in SAR images is designed.The algorithm in this paper can effectively suppress the noise existing in SAR images,and at the same time suppress irrelevant feature channels,thereby effectively utilizing image features.The frequency domain at-tention mechanism achieves denoising function while retaining the effective information of the image,which enhances the robustness of the algorithm.The hybrid pooling mechanism strengthens the road char-acteristics and improves the segmentation accuracy.Real airborne high-resolution SAR image data were used to conduct road segmentation experiments.The results show that the algorithm in this paper has good segmentation effects and its effectiveness is verified.
SAR image road extractionconvolutional neural networkswavelet transformchannel attention
国家科技期刊平台
NSTL
万方数据
