BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//talks.osgeo.org//foss4g-europe-2026//speaker//WFLHZU
BEGIN:VTIMEZONE
TZID:EET
BEGIN:STANDARD
DTSTART:20001029T050000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:EET
TZOFFSETFROM:+0300
TZOFFSETTO:+0200
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:EEST
TZOFFSETFROM:+0200
TZOFFSETTO:+0300
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-foss4g-europe-2026-QYAHU8@talks.osgeo.org
DTSTART;TZID=EET:20260630T153000
DTEND;TZID=EET:20260630T160000
DESCRIPTION:## Motivation\n\nThe proliferation of waste-contaminated areas 
 poses a significant challenge to global ecosystems\, harming wildlife and 
 posing serious risks to human health. Riverine systems are particularly vu
 lnerable\, as floodplains act as temporary storage for mismanaged plastic 
 and debris. During high-water events\, accumulated waste is transported do
 wnstream\, further contaminating aquatic environments.\nGovernmental and n
 on-governmental organizations work extensively to remediate these areas\, 
 but identifying illegal dumpsites along long riverbanks is resource-intens
 ive and often requires field surveys by vehicle or boat. Efficient\, large
 -scale monitoring tools are therefore essential. Recent advances in remote
  sensing and machine learning offer promising solutions. This research aim
 s to develop an automated system for detecting plastic waste along riverba
 nks and water surfaces using multispectral satellite imagery.\n\n\n## Key 
 Related Works\n\nThe field of satellite-based waste detection is rapidly e
 volving. Previous efforts by *Magyar et al. (2023)* laid the foundation fo
 r this study by employing a Random Forest (RF) model on PlanetScope and Se
 ntinel-2 imagery.\n\nOther researchers have utilized different sensors and
  algorithms\; for instance\, *Sakti et al. (2023)* introduced the "Adjuste
 d Plastic Index" to reduce noise from vegetation and buildings in Sentinel
 -2 data\, achieving 88% accuracy on vegetation but facing challenges with 
 spectral similarities between buildings and debris.\n*Lanorte et al. (2017
 )* demonstrated the effectiveness of Support Vector Machines (SVM) for agr
 icultural plastic waste detection using Landsat 8 imagery\, achieving over
 all accuracy up to 94%.\nDeep learning approaches have also been explored.
  *Sun et al. (2023)* utilized high-resolution satellite imagery (0.3m–1m
 ) to achieve a 98% detection rate for various waste types\, significantly 
 reducing the time required for expert manual review. *Torres and Fraternal
 i (2021)* employed a Convolutional Neural Network (CNN) based on the ResNe
 t50 architecture to identify illegal landfills in 20cm resolution orthopho
 tos with an F-score of 88.2%.\nWhile these high-resolution studies show gr
 eat accuracy\, our research focuses on the operational utility of more fre
 quently available multispectral data like PlanetScope to monitor dynamic r
 iver environments.\n\n\n## Methodology\n\n### Data Acquisition and Feature
  Engineering\nThe study utilizes PlanetScope multispectral imagery\, which
  provides four spectral bands (RGB + NIR). To enhance the model's ability 
 to distinguish waste from natural surfaces\, the following spectral indice
 s were calculated:\n - **Plastic Index (PI)**: Leverages the higher reflec
 tance of plastic compared to water in the NIR spectrum.\n - **Normalized D
 ifference Water Index (NDWI)**: Used to delineate water features.\n - **No
 rmalized Difference Vegetation Index (NDVI)** and **Reversed NDVI (RNDVI)*
 *: Used to identify and mask healthy vegetation.\n - **Simple Ratio (SR)**
 : Further assists in vegetation classification.\n\n### Training Dataset\nA
  comprehensive training dataset was compiled\, consisting of 27 million pi
 xels. This dataset includes 29 landfills in Romania — identified via loc
 al registries — and the Kisköre reservoir in Hungary\, which is a known
  site for floating waste accumulation. Every pixel was manually annotated 
 into five categories: *Waste*\, *Water*\, *Pasture/Forest*\, *Bare land*\,
  and *Unknown* (including buildings and roads). To improve accuracy\, high
 -resolution aerial imagery was used to differentiate between plastic waste
  and construction debris.\n\n### Model Development and Optimization\nA Ran
 dom Forest classifier was implemented using the Scikit-Learn library. To m
 anage the large dataset\, the model was optimized by limiting tree depth t
 o 20\, reducing the model size from 14GB to a more manageable 2GB without 
 significantly increasing the false positive rate. Furthermore\, because wa
 ste pixels are vastly outnumbered by other classes\, class weights were ap
 plied to mitigate the high false-negative rates caused by data imbalance.\
 n\n### Advanced Processing Techniques\nSeveral techniques were explored to
  refine performance:\n - **Principal Component Analysis (PCA)**: Applied t
 o reduce parameter dimensions and suppress noise. It was found that three 
 principal components retained 90% of the variance.\n - **Seasonal Separati
 on**: Separate models were trained for *summer* (March–October) and *win
 ter* (November–February) to account for variations in vegetation cover a
 nd atmospheric conditions.\n - **Water Masking**: An algorithm was impleme
 nted to mask areas distant from the river course\, thereby eliminating irr
 elevant false alarms in urban or agricultural areas.\n\n### Interactive We
 b Application\nThe results of our research are integrated into an interact
 ive web application that provides a platform for viewing detected waste lo
 cations. The application automatically downloads and classifies the latest
  satellite imagery for monitored areas. The implementation is open-source 
 and is available on GitHub:\nhttps://github.com/GISLab-ELTE/WasteDetection
 /\n\n## Results and Discussion\n\nThe model was validated using test data 
 from the Drina River\, a site not included in the training set\, featuring
  both land-based dumpsites and floating waste islands. The primary RF mode
 l achieved a Match Rate (True Positive) of 29.32% and a Commission Rate (F
 alse Positive) of 28.13%. While the Omission Rate (False Negative) was hig
 h (70.67%) — largely because the model only classified the core of waste
  islands — this was considered acceptable for operational purposes where
  avoiding false leads for clean-up crews is a priority. The model detects 
 the core regions of waste accumulations while maintaining low false positi
 ves\, which is critical for operational deployment.\n\nPCA integration not
 ably improved noise suppression on water surfaces. The PCA-trained model i
 ncreased the Match Rate to 34.99%\, though at the cost of a higher Commiss
 ion Rate (39.01%). The summer-specific model showed a slight improvement i
 n reliability for summer imagery\, reducing the commission rate to 26.1%. 
 Conversely\, winter detection remains a challenge due to shadows and poor 
 weather conditions\, which hinder spectral accuracy.\n\nOur study contribu
 tes (i) a large annotated dataset\, (ii) an operational RF-based detection
  pipeline\, and (iii) an evaluation of trade-offs between accuracy and usa
 bility in riverine waste monitoring.
DTSTAMP:20260605T011141Z
LOCATION:A01
SUMMARY:Automated Riverine Waste Detection Using Random Forest and Multispe
 ctral Satellite Imagery - Máté Cserép
URL:https://talks.osgeo.org/foss4g-europe-2026/talk/QYAHU8/
END:VEVENT
END:VCALENDAR