Speakers Bio & Abstract

Prakhar Verma Associate Software Engineer
TomTom India

BiographyA Machine Learning enthusiast currently working with TomTom India on devising and developing new innovative technologies to revolutionize map data. I am inquisitive with the desire to improve existing solutions and look for new approaches. I am passionate about technology revolving around neural networks and deep learning and their applications relating to the GIS domain.Abstract Feature Extraction from Satellite ImageryThe proliferation of high resolution satellite imagery has made automated feature detection a reality. This paper proposes a method of detecting and extracting various features from satellite imageries using machine learning capabilities. The proposed framework consists of dataset creation for training machine learning algorithm from existing shapefiles of features, pre-processing satellite images to correct distortions (caused by atmospheric conditions) in data, training a neural network to extract different features, post-processing the result to achieve refined edges and producing a shapefile. The results show that the proposed algorithm is useful for accurately identifying different features for country creation or change detection and saving hundreds of person hours.