Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : Université Bretagne Sud / IRISA
Durée : 3 ans
Contact : charlotte.pelletier@univ-ubs.fr
Date limite de publication : 2024-03-10

Contexte :
During the last decades, the number and characteristics of the imaging sensors on board the satellites have constantly increased and evolved allowing access (often free of charge) to a large amount of Earth observation data. Recent constellations frequently revisit the same regions at a high spatial resolution. For example, the two Sentinel-2 satellites capture all land surfaces every five days at the equator at a 10-meter spatial resolution at best. The data cubes acquired by these
sensors, commonly referred to as satellite image time series (SITS) [8], combine high spectral, spatial, and temporal resolutions, facilitating precise monitoring of landscape dynamics. [1].

The automatic transformation of these data cubes into meaningful information (e.g., deforestation maps or land cover land use maps) usually relies on supervised learning techniques. Recent advances in this field have been marked by a shift towards deep learning methods, owing to their state-of-the-art results across various domains, including computer vision and natural language processing. The ability of temporal neural networks to handle sequential data (e.g., text or audio) and to detect time-invariant characteristics results in various achievements for time series classification in several domains [2], including remote sensing [6].

However, models are often trained statically. In other words, either a model is fine-tuned, leading to forgetting the knowledge gained previously, or a new model is trained for each new dataset
neglecting the opportunity to leverage insights from prior training instances. For example, the French scientific panel on land cover mapping (CES OSO) produces annual land cover maps by
retraining a model every year, overlooking the potential utility of previously trained models. This approach is not only computationally intensive and time-consuming but also suboptimal given the
rapid availability of satellite imagery for model updates. A compelling alternative lies in dynamic learning paradigms, wherein models are updated from a data stream, enabling the accumulation of knowledge over time while mitigating the risk of catastrophic forgetting. In the deep learning era, this strategy is known as continual learning [4]. Traditional scenarios view each observation sequentially and process them independently [10], which is an issue for SITS whose temporal structure (e.g., crop growth rate) is crucial to model landscape dynamics.

Sujet :
While the formal definition of continual learning is much debated in machine learning and computer vision communities, it is non-existent for SITS. The PhD aims at developing for the first time continual learning techniques adapted to the specificities of SITS data by leveraging both continual learning and SITS analysis research. It will consist of two main objectives: (1) devising and evaluating new robust continual learning paradigms for SITS, and (2) refining the continual learning strategy to discover new classes over time. We aim to demonstrate the potential of continual learning applied to SITS for forest monitoring, especially to help monitor Amazon deforestation and degradation on a large scale.

1. Developing new continual learning algorithm for sequences of satellite images. In this regard, we aim to evaluate existing state-of-the-art techniques and their ability to recall dynamics in SITS (e.g., vegetation growth). Among several ideas, we first plan to study continual learning strategies on temporal neural networks (e.g., Transformers with regularized attention weights) when subtime-series are inputted. This scenario requires studying how catastrophic forgetting will impact temporal neural architectures and potentially how its effect can be
mitigated. It also requires to determine the optimal number of past observations to find a trade-off between the precision of the method and memory used to store the data.
2. Discovering new classes over time without forgetting previous class [3]. The inability of continual learning approaches to discover new classes limits considerably their application in real-world remote sensing settings where land cover changes over time and labels cannot be easily collected. A possible idea is to represent each existing class through prototypes, that can be extracted for SITS [9], and maintain them over time. New classes could be identified when embedding of newly acquired observations are dissimilar from existing prototypes.

Profil du candidat :
We are looking for a candidate
• with a computer science, (geo)data science, or statistics master degree (or equivalent),
• with strong data analysis, machine learning, and computer vision knowledge,
• who is familiar with deep learning techniques,
• with excellent programming skills in at least one language (C/C++, Python, etc.),
• with good communication skills (at least in English) are required,
• with interest in Earth observation applications.
• Knowledge of time series analysis and remote sensing techniques will be appreciated

Formation et compétences requises :
Computer science, (geo)data science, or statistics master degree (or equivalent).

Adresse d’emploi :
Université Bretagne Sud
Laboratoire IRISA
Campus de Tohannic
56000 Vannes

Document attaché : 202402091126_PhD_DECOL__CL4SITS.pdf