Estudio de nuevas herramientas para el fenotipado vegetal de alta resolución y sus aplicaciones en agricultura

• Autores: María Victoria Díaz Galián
• Directores de la Tesis: Marcos Egea Gutíerrez-Cortines (dir. tes.) , Pedro J. Navarro Lorente (codir. tes.)
• Lectura: En la Universidad Politécnica de Cartagena ( España ) en 2021
• Idioma: español
• Tribunal Calificador de la Tesis: Héctor Candela Anton (presid.) , Fernando Pérez Sanz (secret.) , María Francisca Rosique Contreras (voc.)
• Enlaces
  • Tesis en acceso abierto en: Repositorio Digital de la UPCT
• Resumen

  • Resumen de la tesis:

    Phenotype is composed by the observable characteristics produced by genotype (set of genetic material) and environment. Phenotyping is the process for determining organism´s phenotype using different technologies and tools, as image analysis with computer vision, statistics, artificial intelligence, among others.

    Farmers have always used phenotyping to select species and to obtain more desirable varieties according to population´s liking. Currently, it is used in laboratory and in open field for this purpose, but also the knowledge obtained after phenotyping can serve as an indicator of alterations. These may be due to internal (mutations or cross-breedings) or external modifications (changes in environmental conditions, such as light and temperature, mainly). External features affect plants because an internal clock, called circadian clock, regulates gene expression according to environmental signals (for instance, day/night or seasonal cycle) allowing plants to adapt and predict new conditions.

    As previously mentioned, the presence of new technologies, subsequently mentioned, opens the possibility to develop new systems for a more rapid and easier plant phenotyping. Furthermore, they do not need the use of professionals, who require a previous knowledge. Then, if the decision making is carried out by a machine, human mistake cannot happen. Consequently, investing in these kinds of gadgets can result cheaper than hiring professionals. In scientific research, these new tools allow to obtain data not described because of the impossibility of getting them previously. As a result, plant behaviour and phenotypic alterations can be better understood.

    Regarding to new techniques, it is worth mentioning the use of new lighting systems, computer vision, and artificial intelligence.

    Lighting modifications allow to alter flowering periods and, consequently, obtaining fruits out of harvest time. Indeed, knowing which photoperiod and wavelengths are involved in a specific plant process, which wants to be activated or inhibited, allows to use supplementary lighting to modify the normal plant behaviour. This is shown in Chapter I where the effect of increasing the photoperiod was studied. Moreover, two different combinations of wavelengths were analysed in order to test which one produced the best production. Strawberry production was enhanced up to 300% in a research greenhouse increasing daylength without losing fruit quality. These studies were also done in a commercial greenhouse. In this case, total production does not change, but amount of first quality fruits increases. Consequently, farmers can obtain more economic profits with the same crop. Therefore, the ability of external conditions to modify plant behaviour and its subsequent applications in agriculture were confirmed.

    Another useful tool is computer vision which acquires images that the human eye is unable to detect using different kinds of cameras, such as hyperspectral, thermal or infrared. Some examples are observing substances, pigments or behaviours not described previously. Although the aim is to automatically process these images, there are cases in which this is highly rough and the user prefers to do it manually or semi-automatically. Other drawback happens when plant behaviour is studied in time series, as showed in Chapter II, where this issue is addressed, as well as how to solve it. In the same way, it is showed how to process data from the images acquired for a more detailed analysis. Before doing this kind of experiment, it is also relevant to know the growth habit and which features want to be studied in order to adjust the vision system to these requirements.

    How to use computer vision to detect changes in plant behaviour is also observed in Chapter III. In this case, flower opening of Wild-type Petunia and Petunia RNAi:PhELF4. It was found a behaviour not described, called flowering in chain, which happened in wild-type organisms. This behaviour seems to be a control system to avoid high energy waste which could be produced by opening lots of flowers at the same speed and time. Therefore, it indicates that flowers should be in contact each others. This mechanism was unknown, so our results uncover a function of ELF4 in coordination of flower opening. Previous studies have already proved that mutants in ELF4 in Arabidopsis produced an increase in production of CONSTANS (CO). Moreover, this behaviour was inhibited in silenced lines confirming ELF4 as a gene involved in flower opening. In addition to this, if an automatic program was created to analyse flower opening, this would allow to do a screening in mutation studies saving time and money. This study also shows the ability of phenotyping in other approaches than selecting new varieties.

    As previously commented, artificial intelligence could be applied in combination with computer vision creating automatic phenotyping tools. In order to achieve automatic programs to study plant phenotype, the availability of ground truth datasets is crucial for algorithm development and testing. Ground truth datasets are sets of images related to real features enabling to calibrate programs or to help during the interpretation.

    Thus, Chapter IV describes the obtention of two ground truth datasets. Images from Antirrhinum flowers and grape berries of different varieties were obtained using two multispectral cameras. These images were acquired to avoid the presence of shines or shadows. In addition to the obtention of these images, additional phenotypic parameters were measured in the laboratory. In flowers, length, width, weight and anthocyanin content were measured. Therefore, the flower development stage could be detected with images using this ground truth dataset by creating an artificial intelligence program.

    In grapes, the features measured were brix degrees and anthocyanin content allowing to create tools to study quality fruit. Thus, new softwares could avoid food fraud or help farmers to know the appropriate moment to harvest in a non-invasive manner.