Practical Part

Applying deep learning techniques for agricultural remote sensing under small data constructions.

Description: The two previous lectures introduced the theoretical foundations and strategies for addressing small-data challenges in remote sensing for agriculture using deep learning. Building on that conceptual foundation, this practical module is designed to help learners transition from theory to hands-on application. Here, we provide step-by-step examples demonstrating how to implement supervised and self-supervised learning models for remote sensing tasks under realistic data constraints using read data from LUCAS dataset. The focus remains on techniques that are relevant to small or weakly labeled datasets, reflecting real-world conditions in agricultural monitoring.

Throughout this practical work, you will:

Train a supervised learning model using a labeled image dataset
Explore and compare self-supervised learning models such as SimSiam, SimCLR, and MoCo
Understand how to extract and visualize feature representations learned by these models
Reflect on the differences in learning behavior between supervised and self-supervised models, particularly in low-data scenarios

Each example is structured to allow adaptation to your own datasets and tasks. By the end of this module, you will not only be familiar with how to run these models but also understand why certain approaches are more effective in small-data settings, reinforcing the practical impact of concepts introduced in the lectures.

Duration: ~90–120 minutes

Practical part flow (content):