Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Plant detection from ultra high resolution remote sensing images: A Semantic Segmentation approach based on fuzzy loss

Shivam Pande, Baki Uzun, Florent Guiotte, Thomas Corpetti, Florian Delerue, Sébastien Lefèvre

TL;DR

This study tackles the challenge of identifying plant species from ultra high resolution (UHR) remote sensing images by introducing a fuzzy loss within the segmentation model, and incorporates Gaussian filter refined GT, introducing stochasticity during training.

Abstract

In this study, we tackle the challenge of identifying plant species from ultra high resolution (UHR) remote sensing images. Our approach involves introducing an RGB remote sensing dataset, characterized by millimeter-level spatial resolution, meticulously curated through several field expeditions across a mountainous region in France covering various landscapes. The task of plant species identification is framed as a semantic segmentation problem for its practical and efficient implementation across vast geographical areas. However, when dealing with segmentation masks, we confront instances where distinguishing boundaries between plant species and their background is challenging. We tackle this issue by introducing a fuzzy loss within the segmentation model. Instead of utilizing one-hot encoded ground truth (GT), our model incorporates Gaussian filter refined GT, introducing stochasticity during training. First experimental results obtained on both our UHR dataset and a public dataset are presented, showing the relevance of the proposed methodology, as well as the need for future improvement.

Plant detection from ultra high resolution remote sensing images: A Semantic Segmentation approach based on fuzzy loss

TL;DR

This study tackles the challenge of identifying plant species from ultra high resolution (UHR) remote sensing images by introducing a fuzzy loss within the segmentation model, and incorporates Gaussian filter refined GT, introducing stochasticity during training.

Abstract

In this study, we tackle the challenge of identifying plant species from ultra high resolution (UHR) remote sensing images. Our approach involves introducing an RGB remote sensing dataset, characterized by millimeter-level spatial resolution, meticulously curated through several field expeditions across a mountainous region in France covering various landscapes. The task of plant species identification is framed as a semantic segmentation problem for its practical and efficient implementation across vast geographical areas. However, when dealing with segmentation masks, we confront instances where distinguishing boundaries between plant species and their background is challenging. We tackle this issue by introducing a fuzzy loss within the segmentation model. Instead of utilizing one-hot encoded ground truth (GT), our model incorporates Gaussian filter refined GT, introducing stochasticity during training. First experimental results obtained on both our UHR dataset and a public dataset are presented, showing the relevance of the proposed methodology, as well as the need for future improvement.
Paper Structure (11 sections, 4 equations, 5 figures, 2 tables)

This paper contains 11 sections, 4 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Dataset
  4. Data preprocessing
  5. Problem definition
  6. Creation of fuzzy labels
  7. Model architecture
  8. Experiments and Preliminary Results
  9. Training protocols and evaluation
  10. Results and discussion
  11. Conclusion

Figures (5)

  • Figure 1: Raw RGB image over Chichoue site (one of the eight sites).
  • Figure 2: SixP dataset with RGB colour composite with overlapping circular ground truth for plants.
  • Figure 3: Schematic of the U-Net based architecture for semantic segmentation.
  • Figure 4: Visual illustration on the SixP dataset: original image (a), probability map (b) and segmentation map (c) in the classical case, Gaussian-convolved groundturh (d), and probability map (e) and segmentation map (f) with our fuzzy loss.
  • Figure 5: Visual illustration on the Weed dataset: original image (a), probability map (b) and segmentation map (c) in the classical case, Gaussian-convolved groundturh (d), and probability map (e) and segmentation map (f) with our fuzzy loss.