Satellite Signal Processing for Forest Detection Using Convolutional Neural Networks

GABRIEL CAFFARATTI; LEONARDO D. EUILLADES; MARTÍN MARCHETTA; PABLO A. EUILLADES; RAYMUNDO FORRADELLAS

Bahía Blanca

Workshop; 2019 XVIII Workshop on Information Processing and Control (RPIC); 2019

Universidad Nacional de SUR, IEEE

Processing of data gathered from remote sensingdevices like satellite and aircraft-based sensors can provideuseful information about important phenomena related to theearth, like volcano shape and activity, glacier and icebergstracking, urban monitoring, forestation changes, among others.Particularly, forestation detection is useful in different problemslike area desertification assessment, forest health analysis, andland flooding simulations. Different techniques have been appliedto problems related to forest analysis based on the satellite data.However, these approaches require human expert intervention tocorrect them in several ways (like adjusting to different typesof vegetation, seasons or geographic locations), which is tediousand costly. In this paper we address these issues by applyingmachine learning algorithms to the forest detection problem. Themain goal of this work is to reduce the workload of experts toproduce such detection models, and to improve their generality tobe suitable for different conditions. The approach was validatedusing Digital Surface Models (DSM), optical and thermal spectralfirms and forest/no-forest masks, obtained from the Shuttle RadarTopography Mission (SRTM), Landsat-8 and JAXA projects onthe Brazilian?s south-east and Argentinian?s center-east regions.