Optimal Deep Transfer Learning-Based Road Extraction for Intelligent Transportation Systems Using High-Resolution Remote Sensing Imagery

Remote sensing (both satellite and terrestrial) is deemed a suitable technique for collecting data on a large scale effectively and accurately that satisfies the demand for intelligent transportation systems (ITS). In recent times, the research works relevant to the use of remote sensing (RS) technologies towards their execution in transportation systems had raised immensely. Road network extraction will be one of the important projects for intelligent transportation systems, real-time updating road networks, and disaster emergency responses. Road extraction related to high-resolution remote sensing images (RSI) was becoming a hot research topic. At present, many studies are relevant to conventional machine learning methods that can be complex and computational due to impervious surfaces namely buildings and roads that are visible in the images. With this motivation, this study progresses a salp swarm optimization with Deep Transfer Learning based Road Extraction (SSODTL-RE) technique for ITS using high-resolution RSIs. The presented SSODTL-RE technique aims to extract the rod utilizing high-resolution RSIs. Further, to generate a collection of deep features from the RS images, the SSODTL-RE approach employs a capsule network (CapsNet). Moreover, the SSODTL-RE technique applies an adaptive neuro-fuzzy inference system (ANFIS) model for the integration of broken roads and improvises efficiency because of road connectivity. To improve the efficacy of the CapsNet model, the SSO is applied for the process of hyperparameter tuning. The stimulation analysis of the SSODTL-RE system was tested on two benchmark datasets. The experimental results stated the betterment of the SSODTL-RE technique over other DL approaches.