In the context of plant tests to assess the quality of new varieties, evaluation procedures are usually performed manually by visual inspection. Advances in the field of computer vision, speeding up acquisition and image processing, motivate the search for high-throughput automated procedures replacing the direct visual inspection. However, these evolutions raise questions on the validation of the quality of the results from these automatic procedures. In this report,we propose to illustrate these questions with a study case on cysts nematodes counting. In this specific framework, we demonstrate the importance to introduce during the evaluation process semi-automatic tools as methodological intermediate step useful to develop statistical analysis of both direct inspection procedure and automated procedure.
Etienne Belin, Mambaye NDiaye, Valérie Grimault, Sophie Perrot, David Rousseau