BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//talks.staging.osgeo.org//foss4g-it-2023//talk//XL9TFZ
BEGIN:VTIMEZONE
TZID:GMT
BEGIN:STANDARD
DTSTART:20001029T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:GMT
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T020000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:BST
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-foss4g-it-2023-XL9TFZ@talks.staging.osgeo.org
DTSTART;TZID=GMT:20230612T164500
DTEND;TZID=GMT:20230612T170000
DESCRIPTION:The increasing use of Uncrewed Aerial Systems (UAS) has opened 
 up new opportunities for ultra-high-resolution (UHR) land cover (LC) class
 ification using optical data with Ground Sampling Distance (GSD) below 10 
 cm. Coastal sand dune ecosystems are difficult to map due to the variabili
 ty of plant species\, making high-resolution vegetation mapping of these a
 reas crucial for analysing vegetation dynamics\, spatial patterns and pred
 icting species diversity. The extreme similarity of vegetation spectral re
 sponses to multispectral sensors\, the small size of the coastal dune plan
 ts (mostly herbaceous)\, and the large amount of data generated are the ma
 in challenges in achieving ultra-high-resolution LC maps of vegetation map
 ping. \nThis work focuses on developing a VHR vegetation cover classificat
 ion model for three areas of the San Rossore National Park in Italy using 
 data collected by UAS (DJI Phantom 4 multispectral) with a multispectral o
 ptical sensor (RGB\, Redge\, NIR). The machine learning model is trained o
 n two phenological-relevant epochs (September 2021 and May 2022) using a s
 ampling scheme that combines UAS flight acquisition and field vegetation s
 urvey data collected at high precision positioning (dual frequency GNSS). 
 A total of 757 herbaceous and shrub species were sampled.\nThe VHR classif
 ication of 12 species and 2 service classes (Debris and Sand) is multitemp
 oral supervised object-oriented (OBIA)\, characterised by spectral feature
 s\, spectral indices\, elevation\, and texture. Three areas of about 5 hec
 tares each were analysed\, one used solely for transferability tests.\nThe
  calibrated multispectral orthomosaics and the Crown Height Model (CHM) we
 re generated with Structure from Motion-based processing. Textural feature
 s based on Haralick co-occurrence matrix and spectral indices were compute
 d\, resulting in a final dataset of 31 features.\nThe semantic segmentatio
 n was performed using eCognition Developer (Trimble)\, based on the Normal
 ised Difference Vegetation Index (NDVI)\, RGB and CHM of May 2022 dataset\
 ,  resulting in 383’200 elements over the three study areas. Imbalanced 
 datasets\, such as the one of this work\, may lead to inaccurate classific
 ation\, so the borderline synthetic minority oversampling technique (SMOTE
 ) was used for oversampling the training dataset.\nThe random forest algor
 ithm was used to classify tree species\, and feature selection based on GI
 NI impurity was conducted to reduce the dimensionality of the input featur
 es (reduced to 19 based on the statistical distribution of impurity).\nTo 
 verify the accuracy of the model\, a primary accuracy measure based on the
  error matrix was calculated\, and the model was cross-validated using a 1
 00-fold stratified cross-validation. The overall accuracy (OA) was found t
 o be 0.77\, with a standard deviation of 0.14. After feature selection\, t
 he OA slightly decreased to 0.76\, but the processing time was improved\, 
 and the standard deviation was reduced to 0.13. The model was then applied
  to an unseen dataset of the transferability-test area\, and the OA decrea
 sed to 0.62.\nIn conclusion\, using UAS and multispectral ad multi-tempora
 l optical data provides a valuable tool for ultra-high-resolution LC mappi
 ng of vegetation in challenging environments such as coastal sand dunes. T
 he developed vegetation cover classification model based on machine learni
 ng algorithms accurately classifies vegetation species and its performance
 s are in line with the literature. Further research is needed to improve t
 he model's accuracy when applied to different datasets and to extend the m
 odel to map other vegetation-dominated dune environments.
DTSTAMP:20260517T224235Z
LOCATION:Sala Videoconferenza @ PoliBa
SUMMARY:Vegetation Cover Classification of Coastal Sand Dune Ecosystems Usi
 ng Ultra-High-Resolution UAS Imagery and Machine Learning Techniques - Ele
 na Belcore\, Melissa Latella\, Marco Piras\, Carlo Camporeale
URL:https://talks.staging.osgeo.org/foss4g-it-2023/talk/XL9TFZ/
END:VEVENT
END:VCALENDAR