Pixel Reconstruction and Edge Detection of Remote Sensing Images by Deep Learning
Hao Shen and Yixin Jing
In order to obtain useful information in remote sensing images, processing of remote sensing images is required. In this paper, an improved residual dense network (RDN), which reduced the complexity of the network by improving the residual dense block (RDB), was designed for pixel reconstruction of remote sensing images, and then edge detection was performed using U2-net in deep learning. Through the experiments on the base dataset and remote sensing image dataset, it was found that the improved RDN had better pixel reconstruction effect than bicubic and Convolutional Neural Network (CNN) methods, with Peak Signal to Noise Ratio (PSNR) value of 30.07 and Structural Similarity Index Measure (SSIM) value of 0.8636 on remote sensing images; the edge detection based on U2-net could extract clear edges. The experimental results prove the reliability of the improved RDN. The improved RDN can be applied in the processing of actual remote sensing images.
Keywords: deep learning, remote sensing image, edge detection, pixel reconstruction