Full text loading...
-
The Convolution Neural Network for automatic objects detection in Earth satellite imagery
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, 17th International Conference on Geoinformatics - Theoretical and Applied Aspects, May 2018, Volume 2018, p.1 - 5
Abstract
Today, the population maps, based on an accurate built-up vector, are highly valuable products on the GIS market. A large amount of the data is the reason for investigating the way of the automatic objects detection on the satellite imagery. The automatic detection can be made by means of the local statistics calculations on the high resolution imagery. This technique emphasizes the textural features of the buildings. It is a rapid and invariant according to the object’s direction and image brightness. But it has a tendency to label objects with similar texture as buildings. In this case study, a two-stage objects detection approach is presented. The Convolution Neural Network (CNN) is a backend for a lot of the computer vision systems for detection and segmentation of objects on the imagery. In this work it was applied to the pre-allocated buildings. The main CNN parameters and estimation methods are considered background for the designing qualitative classification model. The proposed technology was tested on the scattered buildings within the area of interest in the South-East Asia.