Road Network Extraction Using Multi-path Cascade Convolution Neural Network from Remote Sensing Images

Geographic information about the road is an important type of basic geographic information. The extraction of road information plays a crucial role in urban planning, management of traffic, emergency management, and automatic vehicle navigation. With the advancement of remote sensing technology, high-resolution satellite images are becoming more readily available and are improving in quality, making it possible to use these images for accurate road location. The challenge of detecting roads in remote sensing images is complicated by the presence of blurred road edges, sand coverage, and massive terrain objects in complex environments. This paper proposes a deep learning algorithm-based method for extracting road information from remote sensing images to resolve the above problems. The proposed approach consist of two stages, namely pre-processing and segmentation. Initially, satellite images are collected. The collected images may contain noise and blur. This will affect the segmentation output. Thus, we apply an adaptive median filter to the input image before segmentation to eliminate the noise present therein. After pre-processing, the pre-processed image is given to the input of the segmentation process. For segmentation, we introduced a novel optimized multi-path cascade convolution neural network. The proposed multi-path cascade convolution neural network is enhanced using the adaptive artificial jelly optimization (A(2)JO) algorithm. Finally, from the deep learning model, we obtain the segmented road region. The proposed approach is compared with various state-of-the-art approaches in terms of accuracy, precision, sensitivity, F-score, recall, and specificity.