Semantic land cover change detection using harDNet and dual path coronet

Abstract

ABSTRACT Monitoring the spatial and temporal trends of land use land cover (LULC) change is significant for protecting land assets sustainably. However, complicated scenarios in Remote Sensing (RS) images lead to unsatisfactory outcomes, particularly for some uneven and obstructed objects. Furthermore, there is still a hurdle in efficiently extracting and fusing the complex characteristics of RS images to increase the precision of change detection. Therefore, to get better results while detecting changes in LULC, this research implements effective deep learning approaches for segmentation and feature extraction. In the beginning, a HarDNet-based technique is used to efficiently segment the edge details of the land cover data. After that, the segmented information is sent to the Dual Path Coronet to extract spatial and spectral properties and fuse them to create the feature map of the input. After that, it is sent to the Enhanced Change Prediction Module, which generates a change prediction map with precise information. Last but not least, the suggested method has been tested on the three complex remote sensing change detection datasets, Sun Yat-Sen University (SYSU-CD), High-Resolution Semantic Change Detection (HRSCD), and LEVIR-CD. The quantitative and qualitative analysis findings on the datasets reveal that the suggested technique works better than other standard change detection techniques.