Institution/Company:

Université de Polynésie Francaise

Location:

Papetee

 French Polynesia

Job Type:

PhD

Degree Level Required:

Masters

19 days ago Apply now

Description:

Summary of the research project

The objective of this thesis is to propose a methodology to select and measure in a precise, simple and repeatable way the phenotype of Tahitian pearls on a few thousand individuals per lineage and per year, if possible in a non-lethal way on the candidates (or lethal on parent oysters that were used as genetic material in the transplantation process). The pearl phenotypes are color, luster, orientation (iridescence), aragonite deposition rate, surface and shape defects, and nacre thickness.

The developed phenotyping method will be tested on a sample of 200 to 500 pearls representative of the different existing pearl classes. Once validated, our methodology will be used to perform a genetic selection of oysters to improve the quality of the pearls produced.

This project will be carried out in partnership with the French Polynesian Marine Resources Department and local pearl farmers.

 

General context of the project

The Tahitian pearl is a naturally colored cultured pearl that comes from the grafting and rearing in a natural environment of the pearl oyster Pinctada margaritifera. The pearl consists of a deposit of layers of mother-of-pearl (aragonite platelets) around a nucleus, an artificial nucleus inserted into the oyster by a grafter. The Tahitian pearl enjoys an excellent reputation for quality on the international market and is a benchmark against which all pearl production throughout the world is compared.

 

The quality and beauty of a pearl depends on a large number of criteria: the thickness of its nacre, its shape, the quality of its surface, its color and its luster. An international codification and a local codification in Tahiti (defined in 2001) evaluate the quality of pearls according to these criteria. Today, the evaluation of these criteria is generally done by experts.

 

The process is by no means automatic.We already obtained some results in the automatic evaluation of some of these criteria:

  • We proposed a method for the automatic measurement of the thickness of the nacre of Tahitian pearls, for which we have filed a patent
  • We also defined a method for classifying the luster of Tahitian pearls.
  • We also initiated some works on the classification of the color set of a Tahitian pearl.

 

The main objective of this PhD thesis is to build on the results already obtained by completing them with respect to quality criteria that have not yet been fully studied, such as color, defects and shape, in order to move towards a complete automatic phenotyping of a Tahitian pearl.

The results obtained will make it possible to draw up an identity card of a pearl and its quality on all the assessable points.

Responsibilities:

Objectives

The objective of this thesis is to propose a methodology to select and measure in a precise, simple and repeatable way the phenotype of Tahitian pearls on a few thousand individuals per lineage and per year, if possible in a non-lethal way on the candidates (or lethal on parent oysters that were used as genetic material in the transplantation process). The pearl phenotypes are color, luster, orientation (iridescence), aragonite deposition rate, surface and shape defects, and nacre thickness.

The developed phenotyping method will be tested on a sample of 200 to 500 pearls representative of the different existing pearl classes. Once validated, our methodology will be used to perform a genetic selection of oysters to improve the quality of the pearls produced.

This project will be carried out in partnership with the French Polynesian Marine Resources Department and local pearl farmers.

Issues

Traditionally, the phenotype is easier to measure than the genotype. Classical genetics uses the observation of phenotypes to deduce the functions of genes.

The presence of phenotypic variations due to genetic variations is a fundamental element of evolution by natural selection. The selective value (fitness) of an individual results from its life history traits, influenced by the contribution of thousands of traits. Without heritable phenotypic variation, all individuals would have the same selective value and evolution would only be due to chance (genetic drift).

The phenotype can be observed at the different levels of organization of living organisms, we generally retain the three following levels:

  • at the level of molecules: molecular phenotype;
  • at the level of cells: cellular phenotype;
  • at the organism level: macroscopic phenotype.

Our objective is to propose a computer tool coupled with a protocol to measure the macroscopic phenotype of Tahitian pearls that are automatable, accurate and low cost. This will allow us to study genetic crossbreeding of Tahitian pearl oysters to improve the quality of the pearls produced.

Methods

We have already started this work during two previous theses. During this thesis, our objective will be to complete these first results in order to automatically draw a complete phenotype of a Tahitian pearl:

  1. Mother-of-pearl thickness (achieved)
  2. Shell thickness
  3. Color (partially realized)
  4. Luster (partially done)
  5. Orient (iridescence) (partially realized)
  6. Aragonite deposition rate
  7. Surface and shape defects

Concerning the thickness of the nacre (1.), we proposed a complete image processing chain allowing to segment the pearl, then to detect the possible cavities and the nucleus in order to draw a profile of the thickness of the nacre. This method has been patented. A software based on this patent is currently used at the Direction des Ressources Marines de Polynésie.

Concerning the thickness of the shell (2.), we will have to develop a non-invasive measurement system, for example based on ultrasound systems.

Concerning the measurement of pearl color (3.), we have obtained preliminary results during the PhD thesis [1]. One of the objectives of this thesis was to quantify and qualify the color of Tahitian cultured pearls. Indeed, a pearl does not present a single color, but a whole set of colors that can be reflected thanks to the reflection of incident light. Another objective was then to learn the color of a set of pearls to classify them in batches with similar color sets. To learn these colors, it was first necessary to characterize them. We proposed a theoretical method of normalization of the color in the Maxwell triangle, then a solution to split it in order to transform it into a 2D histogram whose values can be synthesized by vectors usable by learning methods with kernels (for example Support Vector Machine). However, this work has not been finalized.

Concerning the measurement of luster and iridescence (4. and 5.), we obtained some results during another PhD thesis on the learning of the ordering of the luster of the pearls. We have set up a protocol for the acquisition of normalized photos of a pearl. We have also identified a set of ten luster features. All these features can be extracted from our acquisitions. We proposed a methodology that allows us to classify the luster of Tahitian pearls using these features. Our method allows not only to classify the pearls but also to order them.

Concerning the rate of aragonite deposition (6.), measurements will have to be carried out with a Scanning electron microscope. We will use a large enough set of grafted pearl oysters of at least 6 months old. We will take them all out of their pearl bags to apply a marker on the surface of the pearls, and then we will put them back in water. The growth of aragonite can then be measured on a set of a few pearls that should be split in half every week or two for at least six months.

Concerning the surface and shape defects (7.), we will carry out acquisitions under different angles and then use image processing based on segmentation methods either by regions or by contours.  We will also learn shape using for example kernel methods such as SVM coupled with descriptors such as Fourrier Mellin or Zernike, or with methods based on deep neural networks. The ultimate goal of this part of the PhD thesis will be to obtain from these measurements and treatments a 3D macroscopic profile of the pearl and its defects.

 

 

Qualifications:

The candidate should have good skills in Python, Matlab, C++ and/or Java programming, as well as in machine learning and image processing.

Knowledge and proficiency in at least one machine learning library, such as Pytorch, TensorFlow and/or Keras, would be a plus.

Additional Information:

This PhD thesis will take place on the campus of the University of French Polynesia in Tahiti.

The start date of this PhD thesis is October 1, 2022 (tentative).

Salary: 1,597€ net per month for 36 months.

To apply, the candidate will have to send to the two PhD thesis directors a letter of motivation, a CV as well as his/her last report cards.

Keywords:

phenotyping

machine learning

image processing

SVM

Tahitian pearl

Posted on: