Abstract
Extracting building footprint from aerial photos and satellite imagery has played a vital role in change detection, urban development , and detect the Agricultural land encroachments. The deep neural networks have feature extraction capability and provide the methods to detect and extract building footprint from Satellite imagery with high accuracy. Image segmentation, is the process by that we try to segment an image into coherent parts with two type of segmentation. Semantic segmentation is a form of segmentation that attempts to segment an image into meaningful parts or predefined class labels. The pixel-wise classification task can help us determine if a pixel be included in a particular object in a dataset. Instance segmentation is semantic segmentation with the distinction of classifying each instance of an object as itself. The convolutional neural networks (CNN) used in instance and semantic segmentation. Nevertheless, one of the main problems of extracting building footprint is that most approaches use high-resolution imagery in sampling training data and inferencing phases, whereas not free public available or available with high cost. Or use semantic segmentation that not applicable with closely situated or connected buildings.
Our proposed approach is extracting building footprint low-resolution satellite imagery using the instance segmentation technique.