Soy culture plays an important role in the national economy, and, for this reason, it is necessary to search for methods and technologies that make research in the field of seeds even more efficient. With this in mind, this project aims, through Digital Image Processing methods and the application of Artificial Intelligence, to analyze and classify soybean seedlings. The main objective is to assist researchers in repetitive and error-prone procedures inherent to to vigor tests and germination experiments, especially in the measurement stages. In addition, it is also desired, with the creation of this tool, to decrease the levels of subjectivity that are present in these analyzes, especially in the seedling classification phase. For this, the work will be divided between the creation of a dataset with images of soy seedlings and the development of two modules, one for seedlings segmentations in terms of their fundamental parts and another for the classification of these seed, both modules using Computer Vision techniques. In order to create the dataset, germination experiments will be carried out according to the rules adopted by the brazilian Seed Analysis Rule (RAS) and the resulting seedlings will be scanned so that the images can be analyzed. This base will be processed and then used for training Neural Network models for semantic segmentation and image classification, both important tasks for collecting the information necessary to calculate the vigor of the seed lots.
News published in Agência FAPESP Newsletter about the scholarship: