PhD, Machine learning and image processing

2/27513
label Burse autorenew 2025-09-29, 17:00 history_edu Silviu Marinescu
The University of French Polynesia (UPF) has an open position for a PhD student in the domain of Machine learning and Image Processing.

The main objective of this thesis is to develop machine learning methods for visual identification using Support Vector Machines (SVM). Considered Applications will be related to the Pacific environment. It will include (but will not be limited to ) the following two applications:

1. Detection, enumeration and morphometric measurements of giant farm clams from pictures.

In French Polynesia, there are many farms that need to estimate the population size of their clams and know the characteristics of the clams at key stages of their breeding. Currently, measures are manually taken by experts in the field. These measures are not very accurate and time-consuming. This application aims to automate these measures by (1) defining a protocol for acquiring data on photos captured by any picture camera and (2) processing the acquired data on a computer.



First step will be to use photographs of a representative sample of a farm population to automatically count clams and compute their morphometric characteristics (i.e size, shape, color). Second step will be to extrapolate from the given samples, the clams population size on the farm and the distribution of the clams according to their morphometric features. Thus, this automated system will enable the farmers on the French Polynesian islands to get information such as the number of the clams on their farms their sizes and colors.




2. Recognition of Pacific fishhooks from photographs.

Archeological excavations in Eastern Polynesia has revealed fishhooks as central in reconstructing the Polynesian past history. In Eastern Polynesia archeologists have used fishhooks recovered from archaeological excavations as chrono-cultural indicators. Fishhooks are considered crucial in retracing fishing techniques, which represent a significant characteristic of ancient Polynesian life. In order to draw conclusions about the fishing techniques used by ancient Polynesians, it is first necessary to classify the fishhooks found in archaeological excavations, and take morphometric measurements. However, undamaged fishhooks are very rare while fragments are much more frequent.

Objective of this application will be threefold. First it will be to automate the classification and the measurement of undamaged fishhooks; second it will be to determine the type of the fishhook from its fragments; and finally it will be to reconstitute the fishhooks from fragments and extrapolate their morphometric measures.