A GF-3 SAR Image Dataset of Road Segmentation

We constructed a GF-3 SAR image dataset based on road segmentation to boost the development of synthetic aperture radar (SAR) technology and make images be applied practice better. selected 23 scenes in Shaanxi, China, cut them into chips with 512 × pixels, then labeled using LabelMe labeling tool. The consists 10026 chips, these are from different imaging modes, so there is diversity resolution polarization. Three algorithms such as Multi-task Network Cascades (MNC), Fully Convolutional Instance-aware Semantic Segmentation (FCIS), Mask Region Neural Networks (Mask R-CNN) trained by dataset. experimental result measures including Average Precision (AP) Intersection over Union (IoU) show that work well this dataset, accuracy R-CNN best, which demonstrates validity we constructed.