Thesis subject

Spectral Reconstruction for Plant Disease Monitoring (MSc)

Plant diseases pose a major and widespread challenge to global food security and agricultural sustainability. Hyperspectral remote sensing has recently become a promising technique for monitoring plant diseases, providing a non-destructive approach that also offers the possibility of early detection and disease identification at a large scale.

Short description

Plant diseases pose a major and widespread challenge to global food security and agricultural sustainability. Hyperspectral remote sensing has recently become a promising technique for monitoring plant diseases, providing a non-destructive approach that also offers the possibility of early detection and disease identification at a large scale. However, the practicality of hyperspectral imaging is hindered by high costs and complex data processing requirements. Addressing these challenges, our project proposes a novel approach: using multispectral data to reconstruct hyperspectral images. This method could revolutionize disease monitoring by making high-resolution spectral analysis more accessible and cost-effective. By leveraging this technique, we aim to enhance disease detection accuracy, leading to more efficient agricultural practices and potentially increasing crop yields.

    Objectives

    This project is connected to the "Detection of the early onset of plant disease by multi-sensor data" project. In comparison, this project has a stronger focus on remote sensing image processing, hyperspectral image reconstruction, and disease simulation model development. Key points include:

    • Remote Sensing Data Processing: This involves standardizing hyperspectral remote sensing data, multispectral data, and simulated data based on radiative transfer models to meet the requirements of hyperspectral reconstruction algorithms. Some data has already undergone radiometric and geometric corrections in the aforementioned project, while simulated data is generated by radiative transfer models.
    • Comparing Hyperspectral Reconstruction Algorithms: We aim to validate the feasibility of reconstructing hyperspectral images from multispectral images. State-of-the-art deep learning models based on RGB spectral reconstruction will be adapted for vegetation spectral reconstruction from multispectral data. The performance of these models will be evaluated using multidimensional image assessment metrics, highlighting their strengths and weaknesses in the spectral reconstruction process, and exploring algorithmic improvements.
    • Developing Disease Monitoring Models: The project will investigate the performance of the reconstructed spectra in plant disease monitoring.

      Tasks

      The work in this master thesis entails:

      • Process various types of remote sensing data.
      • Evaluate and compare different hyperspectral reconstruction algorithms.
      • Develop and test models for plant disease monitoring using reconstructed spectral data.

      Literature

      Requirements

      • Courses: Remote Sensing (GRS-20306), Programming in Python (INF-22306), (Optional), Deep Learning (GRS-34806) or Machine Learning (FTE-35306)
      • Required skills/knowledge: basic data analytics/machine learning and willingness to learn new algorithms.

        Key words: Remote Sensing, Hyperspectral Image, Deep Learning, Transformer

        Contact person(s)