Bachelor's degree theses

NOTE: In order to apply for any of the following theses or internships, the candidate must have no more than 3 exams left.

MACHINE LEARNING

Deep Learning applied to soil parameter retrieval from hyperspectral data

Background
Satellite Remote Sensing is a key technology to foster a more sustainable agriculture. Indeed, it is possible to exploit hyperspectral images to estimate important soil parameters which can, for example, be used to optimize the fertilization process, ultimately reducing the amount of fertilizers needed. Unfortunately, however, the download of hyperspectral images typically requires a lot of transmission power and ground stations’ time.
Remote sensing would likely benefit from deep learning, that, in recent years, has emerged as a possible solution to enhance efficiency and autonomy of space missions at various levels. Onboard image processing through Convolutional Neural Networks (CNN) would significantly reduce the volume of data to be transmitted, also contributing in shortening the amount of time going from image collection to information delivery, to the benefit of the final users.
Thesis topic
The purpose of the thesis work is to develop algorithms, leveraging CNNs, in order to automate the process of estimating soil parameters, such as potassium, phosphorus pentoxide, magnesium, and pH from hyperspectral images.
The thesis work fits within the context of the “Seeing Beyond the Visible” challenge [1] hosted by KP Labs in the framework of the IEEE International Conference on Image Processing (ICIP) 2022. The candidate will therefore be invited to take part in the challenge with the support of a PhD student.
However, the scope of the thesis is not necessarily limited to the participation in the aforementioned competition.
Activities:

  • Literature review on methods to estimate soil parameters from hyperspectral images
  • Crash course on CNNs
  • Identification and implementation of the most promising approach to date
  • Study and implementation of solutions to improve the algorithm
  • Submission of the results to the “Seeing Beyond the Visible” challenge

Requirements

  • Intermediate knowledge of the software MATLAB
  • Willingness to learn the basics of Python and TensorFlow

 

Number of students required: 1 bachelor (internship+thesis)

[1] Seeing Beyond the Visible, https://platform.ai4eo.eu/seeing-beyond-the-visible

Contacts

Prof. Alfredo Locarini

Via Fontanelle 40, 47121 Forlì (FC)

Write an e-mail

Go to the website

Available by appointment

Prof. Dario Modenini

Via Fontanelle 40, 47121 Forlì (FC)

+39 0543 374 450

Write an e-mail

Go to the website

Available by appointment

Prof. Paolo Tortora

Via Fontanelle 40, 47121 Forlì (FC)

+39 0543 374456

Write an e-mail

Go to the website

Available by appointment