Convolutional Neural Network Based Flood Detection Using Remote sensing images

EasyChair Preprint 2235

Abstract

The proposed system uses pre-existing CNN, namely Alex Net, for mapping flooding regions using remote sensing images and the application of high-level spatial features for classification of satellite imagery has been underrepresented. This study aims to address the lack of high-level features by proposing a classification framework based on convolutional neural network (CNN) to learn deep spatial features for flood mapping using optical remote sensing images. Designing a fully trained new convolutional network is impossible due to the limited amount of training data available in most remote sensing studies. Every patch is normalized and resized to feed the network. The designed convolution kernels are then applied to the input patch to generate some normalized features. Each feature highlights a group of similar objects. The kernels are designed to highlight the group of pixels that decrease the loss function. These high-level features mostly rely on the spatial information in the patches. In detection of wetland on flood affected regions using remote sensing images accuracy was employed. The classification results obtained by the deep CNN were compared with those supported based on well-known ensemble classifiers. In this classification scheme is the first attempt, investigating the potential of fine-tuning pre-existing CNN, for cowl mapping of flooding regions. It also serves as a baseline framework for future scientific research using the latest state-of-art machine learning tools for processing remote sensing data.

Keyphrases: Alex Net, Convolutional Neural Network, Flood Detection, High level spatial feature, Kernels

@booklet{EasyChair:2235,
  author    = {C. Thirumarai Selvi and S. Kalieswari},
  title     = {Convolutional Neural Network Based Flood Detection Using Remote sensing images},
  howpublished = {EasyChair Preprint 2235},
  year      = {EasyChair, 2019}}