Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Detection of Bark Beetle Attacks using Hyperspectral PRISMA Data and Few-Shot Learning

Mattia Ferrari, Giancarlo Papitto, Giorgio Deligios, Lorenzo Bruzzone

TL;DR

The paper tackles the challenge of detecting bark beetle infestations in conifer forests when ground-truth data are scarce and satellite hyperspectral resolution is available. It introduces a few-shot learning pipeline that first pre-trains a one-dimensional CNN encoder with contrastive learning (SimCLR) to produce compact spectral embeddings, then uses independent SVRs to estimate pixel-level abundances for Healthy, Affected, and Dead trees from these embeddings. In a Dolomites study using PRISMA data, the approach yields a lower average RMSE (0.1029) than baselines relying on full PRISMA bands (0.1353) or Sentinel-2 bands (0.1770), demonstrating the value of hyperspectral detail combined with few-shot learning for forest health monitoring. The method offers scalable, sub-pixel mapping suitable for large-scale monitoring and suggests potential applicability to other ecological conditions and hyperspectral platforms.

Abstract

Bark beetle infestations represent a serious challenge for maintaining the health of coniferous forests. This paper proposes a few-shot learning approach leveraging contrastive learning to detect bark beetle infestations using satellite PRISMA hyperspectral data. The methodology is based on a contrastive learning framework to pre-train a one-dimensional CNN encoder, enabling the extraction of robust feature representations from hyperspectral data. These extracted features are subsequently utilized as input to support vector regression estimators, one for each class, trained on few labeled samples to estimate the proportions of healthy, attacked by bark beetle, and dead trees for each pixel. Experiments on the area of study in the Dolomites show that our method outperforms the use of original PRISMA spectral bands and of Sentinel-2 data. The results indicate that PRISMA hyperspectral data combined with few-shot learning offers significant advantages for forest health monitoring.

Detection of Bark Beetle Attacks using Hyperspectral PRISMA Data and Few-Shot Learning

TL;DR

The paper tackles the challenge of detecting bark beetle infestations in conifer forests when ground-truth data are scarce and satellite hyperspectral resolution is available. It introduces a few-shot learning pipeline that first pre-trains a one-dimensional CNN encoder with contrastive learning (SimCLR) to produce compact spectral embeddings, then uses independent SVRs to estimate pixel-level abundances for Healthy, Affected, and Dead trees from these embeddings. In a Dolomites study using PRISMA data, the approach yields a lower average RMSE (0.1029) than baselines relying on full PRISMA bands (0.1353) or Sentinel-2 bands (0.1770), demonstrating the value of hyperspectral detail combined with few-shot learning for forest health monitoring. The method offers scalable, sub-pixel mapping suitable for large-scale monitoring and suggests potential applicability to other ecological conditions and hyperspectral platforms.

Abstract

Bark beetle infestations represent a serious challenge for maintaining the health of coniferous forests. This paper proposes a few-shot learning approach leveraging contrastive learning to detect bark beetle infestations using satellite PRISMA hyperspectral data. The methodology is based on a contrastive learning framework to pre-train a one-dimensional CNN encoder, enabling the extraction of robust feature representations from hyperspectral data. These extracted features are subsequently utilized as input to support vector regression estimators, one for each class, trained on few labeled samples to estimate the proportions of healthy, attacked by bark beetle, and dead trees for each pixel. Experiments on the area of study in the Dolomites show that our method outperforms the use of original PRISMA spectral bands and of Sentinel-2 data. The results indicate that PRISMA hyperspectral data combined with few-shot learning offers significant advantages for forest health monitoring.

Paper Structure

This paper contains 4 sections, 4 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Experimental Setup and Results
  4. Conclusions

Figures (3)

  • Figure 1: Scheme of the proposed method.
  • Figure 2: Illustration of the study area location and ground truth data.
  • Figure 3: Portion of the classification maps visualized by normalizing each individual RGB channel, corresponding to the three considered classes, to the minimum and maximum values across the maps. The resulting colors reflect the varying abundances of each of the three classes for every pixel.