Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Super-Resolution Analysis for Landfill Waste Classification

Matias Molina, Rita P. Ribeiro, Bruno Veloso, João Gama

TL;DR

This work tackles illegal landfill detection by bridging the gap between high-resolution labelled datasets and open-access low-resolution imagery through cross-domain waste classification and super-resolution. It trains a ResNet-50 classifier on high-resolution data and systematically evaluates performance across downscaled resolutions, introducing a third experiment that uses an SR model (EDSR) to upscale low-resolution inputs before classification. The study finds that classification performance declines with lower resolutions, but super-resolution preprocessing can recover much of this loss while increasing model sensitivity, necessitating threshold tuning. The results provide a framework for applying landfill detection in practical, resource-constrained settings and highlight the importance of threshold customization to domain needs.

Abstract

Illegal landfills are a critical issue due to their environmental, economic, and public health impacts. This study leverages aerial imagery for environmental crime monitoring. While advances in artificial intelligence and computer vision hold promise, the challenge lies in training models with high-resolution literature datasets and adapting them to open-access low-resolution images. Considering the substantial quality differences and limited annotation, this research explores the adaptability of models across these domains. Motivated by the necessity for a comprehensive evaluation of waste detection algorithms, it advocates cross-domain classification and super-resolution enhancement to analyze the impact of different image resolutions on waste classification as an evaluation to combat the proliferation of illegal landfills. We observed performance improvements by enhancing image quality but noted an influence on model sensitivity, necessitating careful threshold fine-tuning.

Super-Resolution Analysis for Landfill Waste Classification

TL;DR

This work tackles illegal landfill detection by bridging the gap between high-resolution labelled datasets and open-access low-resolution imagery through cross-domain waste classification and super-resolution. It trains a ResNet-50 classifier on high-resolution data and systematically evaluates performance across downscaled resolutions, introducing a third experiment that uses an SR model (EDSR) to upscale low-resolution inputs before classification. The study finds that classification performance declines with lower resolutions, but super-resolution preprocessing can recover much of this loss while increasing model sensitivity, necessitating threshold tuning. The results provide a framework for applying landfill detection in practical, resource-constrained settings and highlight the importance of threshold customization to domain needs.

Abstract

Illegal landfills are a critical issue due to their environmental, economic, and public health impacts. This study leverages aerial imagery for environmental crime monitoring. While advances in artificial intelligence and computer vision hold promise, the challenge lies in training models with high-resolution literature datasets and adapting them to open-access low-resolution images. Considering the substantial quality differences and limited annotation, this research explores the adaptability of models across these domains. Motivated by the necessity for a comprehensive evaluation of waste detection algorithms, it advocates cross-domain classification and super-resolution enhancement to analyze the impact of different image resolutions on waste classification as an evaluation to combat the proliferation of illegal landfills. We observed performance improvements by enhancing image quality but noted an influence on model sensitivity, necessitating careful threshold fine-tuning.
Paper Structure (13 sections, 5 figures, 4 tables)

This paper contains 13 sections, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Image classification for landfills discovery
  4. Image quality improvement
  5. Methodology
  6. Experimental Setup
  7. Metrics.
  8. Experiment I.
  9. Experiment II.
  10. Experiment III.
  11. Training and evaluation.
  12. Results
  13. Conclusions

Figures (5)

  • Figure 1: Super-Resolution (SR) example: low-resolution input (left), SR output (center) and high-resolution ground truth (right).
  • Figure 2: Evaluation of models trained on different resolutions (Experiment I) - solid lines - and on maximum resolution (Experiment II) - dashed lines - by image size.
  • Figure 3: Evaluation of models trained on different resolutions (Experiment I) - solid lines - and SR enhancement with default threshold (Experiment III) - dashed lines - by image size.
  • Figure 4: Evaluation of models trained on different resolutions (Experiment I) - solid lines - and SR enhancement with best threshold (Experiment III) - dashed lines - by image size.
  • Figure 5: ROC curves obtained with SR enhancement (Experiment III).