Feature Detection for Satellite Imagery Using Deep-Convolutional Neural Network

Abstract

This paper deals with the implementation approach for the feature detection for the images obtained from satellites using deep convolutional neural network. The aim of this paper is to accurately segment different images of different classes of the satellite imagery. Our method for the implementation is confined to the adaptation of CNN for multispectral data processing. We have also stated and developed several changes for the training objective and the training pipeline our method of implementation of the above model secured 3rd place from 419 entries that were submitted. The accuracy of the model implemented can be compared to the methods that secured the first 2 places in the competition, but unlikely this satellite image analysis does not depend on the complex assembling schemes so can be utilized in the automatic featuring at ease.