Desarrollo de un sistema de fenotipado basado en visión artificial para el estudio de la cinética de crecimiento en plantas

Résumé

The growing world population is causing the need to increase crop production. Scientists and farmers are seeking to obtain genetic varieties that are more productive, stress resistant and less demanding in terms of resource consumption. The technological advances of recent decades have made it pos- sible to obtain a multitude of new genetic lines rapidly and economically. It is also necessary to validate these new lines and check whether their phenotypic behaviour complies with what is expected according to the genotype created. However, the bottleneck of this process occurs precisely in the phenotyping of the lines. The rate of generation of new varieties has been greatly outpacing our ability to identify their phenotypes. In recent years, the development of the High Throughput Phenotyping (HTP) has been able to shorten these distances, due to the growing interest shown by corporations and governments. Computer vision-based phenotyping is a noninvasive method in most cases. It has a multitude of possible configurations depending on the objectives set. It is becoming more affordable thanks to the lower cost of technology and is multiscale, and can be implemented from a small cultivation chamber to large agricultural areas. Therefore, this work proposes the development of a phenotyping system based on computer vision, to be implemented in a growth chamber, with them main aim of studying the growth rate of different genetic lines –wild type and transgenic–. This system not only includes devices and the development of control software, but also the implementation of image processing methodologies and growth analysis. Thus, in this thesis we present a method of segmentation of leaves over substrate background, by creating feature vectors based on color spaces –Red Green Blue– or Wavelet transforms -Near Infrared –, as a basis for training machine learning classifiers. In both cases, classification yields of close to 100 for the set oftested images. On the other hand, the application to the study of the growth of different lines of Antirrhinum majus and Petunia x hybrida, has allowed the implementation of a modelling methodology –Generalized Additive Models–, which facilitates the study of the growth curves and allows the determination of parameters such as maximum and mean growth rates. The application of phenotyping systems based on artificial vision, shows a great potential for the study of uncharacterized germplams, as a result of genomic and epigenetic changes produced in plants. Finally, it would be desirable to focus economic and human efforts towards a field that can provide not only scientific knowledge, but also solutions for the agriculture and food industry.