Due toitsuniquegeography,theriversoftheAmazonplayasignificant roleinthedailylife
of theregion,includingbeingtheonlyaccessinmoreremotecommunities,andthelackof
infrastructure inmuchofthiswaterwaynetworkmakesitconducivetothetransportofillegal
goods duetothelackofsupervision.
When comparedtourbanareas,itispossibletonoticethepotentialforpositiveassistanceto
local authoritiesthroughtheuseofrecognitiontechnologies,butthelackofcommunication
structures hinderstheimplementationofthesetechnologiesinthesamewayitisdoneincities.
This projectproposesthedevelopmentofaneuralnetworkapplicationaimedatassistingin
public securityissuesthroughthedetectionofvesselsinportregionslocatedintheStateof
Amazonas, combiningArtificial IntelligencetechniquesandEmbeddedSystemsdevices,without
the useofnetworkdatatransmission.Themethodologyconsistsof:(a)Acquisitionofvessel
images atMarinadoDaviandtheportofManausModerna-StateofAmazonas;(b)Preparation
of thedatasetintheexpectedformatforthemodel;(c)Installation,configuration, anduseofthe
Jetson NanoBoardintrainingtheNeuralNetwork;(d)UseofTransferLearning,leveraginga
pre-trained model(YOLOv4Tiny)formodelgenerationonthenewdataset;(e)Validationofthe
obtained results.
The resultsshowthatwiththeuseofversionsoptimizedforembeddeddevices,itwaspossible
to executetheYOLOv4Tinyalgorithm,performingtheclassification ofvesselsandcontributing
with newvesselimages,whichprovidegreaterregionalitytothepresenteddata.
|
