In recent times, technique has gradually become inevitable in human lives because of growth of some areas such as Internet of Things (IoT), big data, and deep learning (DL). In this research, proposed three-layered edge-fog-cloud-depended satellite IoT structure utilized superpixel-based regularized–convolutional neural network (R-CNN) method. In layer of fog, superpixel-depended simple linear iterative cluster (SLIC) method utilizes images captured through edge stage for generating low-sized superpixel images which is transmitted in less bandwidth. The R-CNN method in cloud stage is next trained by superpixel images for predicting road networks from remote sensing (RS) images. The DeepGlobe and Massachusetts road datasets are utilized for evaluating proposed R-CNN method. The proposed R-CNN method attained precision 92.34%, recall 91.76%, f1-score 91.89%, and intersection of union (IoU) 84.35% on DeepGlobe road dataset and then attained precision 90.41%, recall 90.18%, f1-score 90.22%, and IoU 90.77% on Massachusetts road dataset which is efficient than UNet with attention block.