deniseiras / portfolio_pivot-detection

Pivot Detection using CNN and Landsat 8 bands

Pivot-Detection

Identification of pivots using band compositions and a fast Deep Learning method

This repository contains the my work from the paper

Identificação de pivôs centrais usando composições de bandas e um método rápido de Deep Learning EIRAS, D. M. A.;RUIZ, I. H.; SIMÕES, P. S.; SILVA, G. M.; DUTRA, A. C.; SHIMABUKURO, Y. E.; FONSECA, L. M. G.; GALVÃO, L. S. Proceedings XXI GEOINFO, November 30 - December 03, 2020, São José dos Campos, SP, Brazil. p 180-185, 2020.

http://www.geoinfo.info/geoinfo_series.htm#PastEditions

http://urlib.net/rep/8JMKD3MGPDW34P/43PR2H2

ABSTRACT

This paper presents a technique to identify central pivot (CP) using patches of images containing only one CP, composed by varied bands of the Landsat 8 OLI sensor and spectral indices, through a fast Convolutional Neural Network (CNN). Different combinations of bands and indexes were tested in this work, as infrared band and NDVI. The obtained results indicate best accuracy (95,85%) when using bands not commonly used in CNNs, surpassing some works. CNN also demonstrated advantages in terms of speed, by classifying a patch of image containing CP in only 0.28 milliseconds, revealing great potential for CP identification in remote sensing images, available in official catalogs.

METHODOLOGY