Classification and Pixel-Based Segmentation to Evaluate Soybean Seeds Submitted to Tetrazolium Test

Abstract

Production and use of high quality seeds are essential for the soybean crop. Thus, the quality control system in seeds industry must be reliable, precise, and fast. Tetrazolium test evaluates not only seeds viability but also their vigor, as well as provides information concerning agents that cause their quality reduction. Although this test does not use expensive devices and reagents, it requires a well-trained analyst. Its precision depends on knowledge of all techniques and required procedures. Besides, also necessary is the observer’s subjectivity. So, this trial aimed at developing a computational tool that could minimize the implicit subjectivity in carrying out this test. It also contributes to generate a greater credibility of information and to guarantee precise answers. Algorithms of supervised classification were applied based on extraction of digital images characterization of tetrazolium test. This procedure aimed at producing pixel-based segmentation of those images, to produce a digital segmented image of tetrazolium test according to damage classes. This tool allows, based on image of tetrazolium test, to identify damage on soybean embryos, as well as its site and extension on tissues, so that the interpretation is less subjective. The applied method allowed identifying damage on images of tetrazolium tests in a straightforward way, as well as extracting safer information about those damages and carrying out management control of tetrazolium test according to a seed data file.