BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//talks.osgeo.org//foss4g-2022-academic-track//speaker//3V
 LHAA
BEGIN:VTIMEZONE
TZID:CET
BEGIN:STANDARD
DTSTART:20001029T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-foss4g-2022-academic-track-DTQGKM@talks.osgeo.org
DTSTART;TZID=CET:20220826T093000
DTEND;TZID=CET:20220826T100000
DESCRIPTION:The submerged topography of rivers is a crucial variable in flu
 vial processes and hydrodynamics models. Fluvial bathymetry is traditional
 ly realised through echo sounders embedded on vessels or total stations an
 d GNSS receivers whether the surveyed riverbeds are small streams or dry. 
 Besides being time-consuming and often spatially limited\, traditional riv
 erine bathymetry is strongly constrained by currents and deep waters. In s
 uch a scenario\, remote sensing techniques have progressively complemented
  traditional bathymetry providing high-resolution information. To date\, t
 he peak of innovation for bathymetry has been reached with the use of opti
 cal sensors on uncrewed aerial vehicles (UAV) systems\, along with green l
 idars (Vélez-Nicolás et al.\, 2021). The main obstacle in optical-derive
 d bathymetry is the refraction of the light passing the atmosphere-water i
 nterface. The refraction distorts the photogrammetric scene reconstruction
 \, causing in-water measures to be underestimated (i.e.\, shallower than r
 eality). To correct these distortions\, radiometric-based methods are freq
 uently applied. They are focused on the spectral response of the means cro
 ssed by the light and are typically built on the theory that the total rad
 iative energy reflected by the water column is function of the water depth
  (Makboul et al.\, 2017). The primary goal of the research on submerged to
 pography is to understand the relationship between the water column reflec
 tance and the water depth using statistical and trigonometrical models. Th
 e spread of artificial intelligence has given a new light of interest on s
 pectral-based bathymetry by investigating the non-linear and very complex 
 relationship between variables (Mandlburger et al.\, 2021). To train artif
 icial intelligence models\, large amounts of data are usually necessary\; 
 therefore\, participatory approach and data sharing are required to build 
 statistically-relevant datasets. In this scenario\, FOSS tools and distrib
 uted resources are mandatory to manage the dataset and allow the replicabi
 lity of the methodology.  \nThis work aims to test the effectiveness of ar
 tificial intelligence to correct water refraction in shallow inland water 
 using very high-resolution images collected by Unmanned Aerial Vehicles (U
 AV) and processed through a total FOSS workflow. The tests focus on using 
 synthetic information extracted from the visible component of the electrom
 agnetic spectrum. An artificial neural network is created with the data fr
 om three different case studies placed in west-north Italy\, and geologica
 lly and morphologically similar.\nThe data for the analysis were collected
  in 2020. Each data collection was realised using a UAV commercial solutio
 n (DJI Phantom 4 Pro)\, and the following datasets were generated: i) RGB 
 georeferenced orthomosaic of the riverbed and banks obtained from photogra
 mmetric process\, ii) georeferenced Digital Elevation Model (DEM) of the r
 iverbed obtained from photogrammetric process\, iii) GNSS measures of the 
 riverbed and the riverbanks.\nThe UAV-collected frames were elaborated thr
 ough a standard structure from motion (SfM) procedure. Visual SfM was empl
 oyed to align images and the 3D point cloud computation. The digital surfa
 ce model (DSM) and the orthomosaic production were generated starting from
  the point cloud in Cloud Compare software. By applying the so-called dire
 ct-photogrammetry\, the point clouds were directly georeferenced in the WG
 S84-UTM32 coordinate system thanks to the positioning information retrieve
 d from the embedded GNSS dual-frequency receiver (Chiabrando\, Lingua and 
 Piras\, 2013). Using the information regarding the camera position and the
  local height model provided by the national military Geographic Institute
  (IGM)\, the ellipsoidal heights were translated into orthometric heights.
  The GNSS measures had 3 cm accuracy on the vertical component and 1.5cm o
 n the horizontal components. \nThe RGB information\, DSM and seven radiome
 tric indices (i.e.\, Normalised Difference Turbidity Index\; Red and Green
  Ratio\; Red and Blue Ratio\; Green and Red Ratio\; Green and Blue Ratio\;
  Blue and Red Ratio\; Blue and Green Ratio) were calculated and stacked in
  an 11-bands raster (input raster). The Up component of the bathymetry cro
 ss-sections constituted the so-called "Z_GNSS" dataset and is the dependen
 t variable of the regression. The position (Easting\, Northing\, Up) of ea
 ch Z-GNSS observation was used to extract the pixel values of each band of
  the input photogrammetric dataset\, including the photogrammetric DEM. Th
 e dataset was then normalised and divided into test (20% observations) and
  training (80% observations) datasets.\nIn this work\, a 5-layer multilaye
 r perceptron (MLP) networks model with three hidden layers was built in Py
 thon using the deep learning library Keras with TensorFlow backend (Abadi 
 et al.\, 2016). The ReLu activation function was added to the ANN layers t
 o bring non-linear properties in the network. The dimension of the input l
 ayer is 11\, and the weights are initialised to small Gaussian random valu
 es (kernel initialiser 'Normal') despite usually skewed or bimodal. A kern
 el regulizer\, L1\, was added to reduce the overfitting. The applied optim
 iser to update weights in the network is the Adaptive Moment Estimation (A
 dam) search technique\, and the loss function\, which evaluates the model 
 used by the optimiser to navigate the weights\, is the mean absolute error
  between the predicted output and the target output.\nThe network was trai
 ned on the normalised dataset. The r-squared score\, the Mean squared erro
 r and the Mean absolute error were computed. Finally\, the permutation imp
 ortance was measured using the eli5 python library. \nThe neural network r
 egressor performed over 0.80 of r-squared score on the test dataset. As ex
 pected\, the permutation importance analysis reveals the high impact of th
 e DEM and visible bands\, and low importance scores are reported for ratio
 s bands. \nThe results are satisfying and quite relevant\, although the mo
 del is the first step through a more complex and deeper neural network to 
 correct water distortions in rivers. It has been trained on a relatively s
 mall dataset\, but we intend to follow up with the research\, add more dat
 a\, and develop a free and open tool for the scientific community.  The pr
 esent work\, provide a good insight about the high reliability and accurac
 y of artificial intelligence approaches in optical-derived bathymetry.
DTSTAMP:20260616T025544Z
LOCATION:Room Hall 3A
SUMMARY:Laying the foundation for an artificial neural network for photogra
 mmetric riverine bathymetry - Elena Belcore
URL:https://talks.osgeo.org/foss4g-2022-academic-track/talk/DTQGKM/
END:VEVENT
END:VCALENDAR